> In the real world, it does feel likely that we’re going to hit some sort of a ceiling on the number of useful bugs, and probably we’ll hit it soon.
This doesn't resonate with me. I see companies adding more sloppily written features with AI. I see more bugs in the software I use, not less. While it's plausible that software is getting both buggier and more secure, I suspect those two move in the same direction not opposite.
My guess is that we're getting better at finding _existing_ security issues with AI (and thus fixing those issues), but simultaneously adding more insecure surface areas _at a faster rate_.
One way to resolve the tension here is to note that CNE and lawful-intercept access to phones depends generally on platform vulnerabilities, not application code vulnerabilities. Low-level platform code churns less, absorbs more fixes under AI workloads than it does new features, and works in a constrained space where guardrails are easier to provide (and where those guardrails already have institutional support at Apple and Google).
Over the long term this state of play could change, and IC/LEO organizations could start leaning more on application vulnerabilities than on platform RCEs. But the action would probably still coalesce around a couple of app-layer targets that could themselves be hardened.
Disagree, and in a way it feels like we are dealing with inverse issues: the security "skill" is well defined and will be also engaged with by an agent. Communication companies are further incentivized as any failure is at best reputational harm. Meanwhile SaaS companies are not strictly required to have good UX for their human end users, largely because those users will likely work around the issue. also network effect vs low costs of switching for for comms
I don't have any colleagues like that anymore, but even as far back as the last half of 2025 I was seeing that automated AI review was becoming effective enough that I considered it essential to any project where security was a serious concern.
There days we're generating multiple times more code than we were writing before. That means a similar multiple of opportunities for bugs to be introduced - so the ability to automate security review is more impactful in proportion to that.
On one side, you have pieces like this, where seemingly there are constant fights between serious actors with large and properly distributed budgets, employing top tech and top minds; on the other - regular news of the hackz, where responsible person in charge of security with root access failed to grasp basic technical knowledge (several times), ticking every checkbox in "never do this" list from security best practices, which led to every customer being pwned.
It's like two parallel worlds, that exist in the same place at the same time, but somehow don't cross.
I started my career in the military, and got lucky enough that SOMEHOW, we convinced them to fund a trip to GDQ for educational purposes.
Anyways, while there I attended a little roundtable on software security. It was me, representing a small unit from the Air Force, some dude from Google, and like 15 game devs.
Despite only being a dev of 5 years at that point, I was SHOCKED at the lack of knowledge on software security. Even simple concepts seemed completely foreign to the game devs, though the Google dude seemed to have a really solid understanding of security.
Obviously game devs and website devs and all kinds of devs have different focuses, but it just blew my mind that out of all the topics there I might’ve been considered a comparative expert in, security was somehow the one. I wasn’t sure if that was a major plus for military devs or a major concern for the other devs, but now I’m starting to learn in the latter direction.
Without knowing when that was, I can't be sure how concerning it is.
As an end user, security in game development seems to hit a minimum is the early 2000s.
I recall playing EverQuest and using ShowEQ on a Linux machine on the same network as a proxy to silently eavesdrop on the traffic to show a map of the game with all the hidden/obscured/visible stuff all present. They either never encrypted that traffic or I quit playing before they did.
As multiplayer games started to become the default, netcode improved quite a bit and basic encryption was implemented - but I don't recall a single multi-player game, ever, that was free of hackers when playing online.
It feels like over the years the issue moved from remote exploitation to hiding processes on the user's machine. That has lead to kernel-level anticheat systems. I'm unsure of the effectiveness as when they started to become common I moved all of my gaming to dedicated hosts because I wasn't willing to give that level of access to my "real" computers. That said, I've certainly not noticed a decrease in the perceived number of cheaters.
These days, as far as I know - and to be clear, this isn't an area of deep interest for me - I'm not worried about my gaming accounts being compromised through technical means. It feels like that sort of security is a solved issue overall, although I suppose it's also possible that other forms of exploits have simply gotten easier/cheaper.
> This is not a call to action for experts to rally behind a sophisticated plan. Like so many things about the AI revolution, it’s just occurring to me that we’re on a long greasy slide to a place that will look different than where we are today.
heh. long greasy slide. It really does feel like that.
> In fact, the worst part about this dynamic is that these potential new backdoors will begin primarily useful for allowing the US to weaken its own systems, which will in turn allow foreign adversaries to find new ways to attack our communications. This deliberate self-sabotage will happen just at a moment when we’re finally learning how to defend our own infrastructure.
I don't understand how you can both argue for that law enforcement (and intelligence) agencies will force others to implement intentional backdoors AND also everyone will be using AI to find and secure ALL potential holes in the software so there won't be any vulnerabilities anymore.
Wouldn't one AI or another detect this deliberate backdoor and report it, as it'll look just like any other security vulnerability, the only difference being the intention?
I have respect for the author so I feel like I probably misunderstand something from the overall text rather than I somehow have a better perspective on this topic that the author knows very much more about than me. I felt like I nodded along all up until "So how is this a problem?" and now I'm not sure I understood correctly.
I suspect they want a backdoor that basically acts like a front door (current password regimes). That is, a kind of high level password that decrypts traffic given a specific, changing, password that only the government has access to.
It's seems like an odd-duck for sure, and I doubt it's a realistic proposition. I do think "perfect encryption for dummies" is all well and good until organized crime organizations are able to challenge the government in certain regions of our country.
All of this sends us deep into the realm of political philosophy, the nature and purpose of governments, and the freedom vs security tradeoffs we live with.
I listened to the latest Plain English podcast this morning, which was explicitly about the potential for a ransomwarepocalypse in the coming years, as open models let any tom, dick, or harry become capable of a plug-and-play ransomware attack, instead of that being left to the realm of professionals. It's a bit nerveracking to think about every nigerian prince scam suddenly becoming a sophisticated attack on your local water sanitation system.
Ah, like new regulation requiring a "Portal For Law Enforcement" for platforms over X MAU or something?
I think I can kind of see that from the text re-reading it, it's kind of hidden though and not so explicit if this is what he is trying to communicate. Thanks for giving me a new perspective to read it with!
> In fact, the worst part about this dynamic is that these potential new backdoors will begin primarily useful for allowing the US to weaken its own systems, which will in turn allow foreign adversaries to find new ways to attack our communications.
The author agrees with you and addressed this point. The US forcing backdoors into its own systems ("own": Those for sale and distribution within the US) would create vulnerabilities making those systems weaker.
> Wouldn't one AI or another detect this deliberate backdoor and report it, as it'll look just like any other security vulnerability, the only difference being the intention?
That's precisely the author's point: deliberate backdoors will be more adversary-exploitable than ever before, but the demand for such from law enforcement agencies is likely to ratchet upwards.
But that's one step after, I'm trying to understand how those backdoors even end up in software if everything gets automatically reviewed by the people working on these codebases? Wouldn't things like these be flagged by systems other than the developers tasked by the agency to implement it? How can law enforcement actually get these things implemented without big parts of the engineering team not seeing warnings about it happening?
I'm not sure you understand what is being described here?
If you're a US company building an app/device/etc. such that an intelligence agency like the CIA or FBI would want access to the data in that product which is normally secured, then they're not going to try to sneak it in there without your development team knowing. They're going to have a meeting with the owners of the company and say, "hey, we'd really like you to implement this backdoor for us, and in return we won't cause you in problems."
Note that this does certainly already happen a lot, but it's also not something that can happen across the board (like the author points out). Apple, being one of the largest companies in the world and who has one of their biggest selling points being their security, has explicitly refused to do this to the point that the intelligence agencies couldn't break into an iPhone until another company found a way to do so.
>I'm trying to understand how those backdoors even end up in software if everything gets automatically reviewed by the people working on these codebases?
The people working on these codebases are "in on it." Of course, we're talking backdoors which are very subtle, target very minimal infrastructure, and are known about by very few people. But, like I said, companies are currently shipping products with backdoors in them knowing they exist already. AI doesn't change that dynamic.
>Wouldn't things like these be flagged by systems other than the developers tasked by the agency to implement it?
The only people who would have access to the systems that can even be flagged by this stuff would be people who would know about it. Keep in mind that the scale we're talking here is massive. Think about how software development works at companies like Apple, Microsoft, Google, etc. There are devs working in offices all around the world where they only ever have access to a fraction of the code that company owns. These companies are very capable of keeping their stuff locked down. It's a necessary component of their work.
>How can law enforcement actually get these things implemented without big parts of the engineering team not seeing warnings about it happening?
Hopefully my explanation at this point is clear, but just to be concrete: backdoors are, by design, very hard to detect. That doesn't mean they're just sneakily written code that humans don't notice as they read over it, but, instead, they're very subtle implementations in very specific parts of huge systems that are already locked down to the point that the number of people who even have access to those portions of the systems are very limited. These agencies don't slip in backdoors without anybody noticing; they convince the minimal number of people needed to know about it to implement it. Again, we're talking about a meeting between the directory of a three letter agency and a CEO, where the CEO then directs the CTO to implement the backdoor who then instructs the handful of very high-ranking engineers to do so.
AI systems in these companies may very well flag these backdoors to the people who already know they exist, then these people can tell the AI "hey, those are their on purpose, so just move on," and the other 99.99% of the company will never know they exist.
I suppose it's important to emphasize, again, that these systems are incredibly massive and complex and most people at these orgs don't have any access to most of these systems, so it's not like you can expect an intern running BugBot across a repo and expecting it to find a backdoor.
"backdoors" aren't holes, they are well-documented superuser APIs. The hack will be by compromising the API user's credentials (moles, stolen passwords, etc), not compromising the server's design intent.
Man I thought from the title this was going to be about next-gen AI being able to zero day everything so effectively that software security is meaningless and we'd need to basically shut it all down, go dark.
I don't think the thesis that a government will be able to do something will ultimately hold. I don't see how they can avoid "going dark" in a democracy.
we live in a world where the government can't even do much about illegal drug markets anyone can access by downloading a piece of software.
if they pass laws that mandate backdoor access and block software which doesn't conform more and more people will move to the dark networks.
and if they effectively block the dark networks (in the limit they will have to block all encrypted communications) then we will be living in a tyranny.
freedom is messy. accept that digital crime can only be solved when the criminal makes a tangible mistake. LLM's will be building profiles on criminals to help with identifying mistakes.
The problem that Matthew Green is calling out subtextually is that democracy is likely to disappoint nerds on this issue: a random voter in the country 10 years from now may very well not share your priors about this issue.
consider the implications of a government where setting up an encrypted network is a serious offense, or where detecting encrypted communications is routine. the average voter doesn't matter. if they are willing to go that far it's a tyranny and the average voter doesn't matter anyway because they spy on everyone and have LLM's profile everyone (in that context even if voting still happens it doesn't matter, they will nudge the voters however they wish due to the information asymmetry).
The reason why this might worry people who don't like the US/UK/Israeli governments is hidden in a secret place...a paragraph that is neither the first paragraph nor the last.
Tl;dr the argument is if three letter agencies can’t buy or make exploits anymore because all vulnerabilities are patched, they will ramp up legal and legislative pressure to make vendors install backdoors. Which is bad and we should all worry about.
One of the author’s blind spots here is the concept of “digital sovereignty”. The US is continuing to ban more and more Chinese-made hardware out of fear that the Chinese govt has installed backdoors in them… which you could interpret as an admission that the US does exactly that with American made products. Globalized supply chains are, erm, complex, and few if any companies are really going to be able to achieve “digital sovereignty” with hardware. But with software it’s actually plausible, though obviously hard. Whether or not the feds have actually installed backdoors in Microsoft Outlook, foreign govts are rightly concerned that they have, and are increasingly pushing to avoid US made software for simple national security reasons.
Which is to say that they will become more authoritarian and aggressive? That's not a good reason to be unhappy that they can no longer hack shit. They shouldn't do that either.
There are lots of things governments should and shouldn’t do, but unfortunately we don’t live in a vacuum, and need to consider the consequences of actions, rather than simply hope everyone is somehow forced to react the way we expect and desire.
In a nutshell, what I’m saying is “They shouldn’t do either” is at best a platitude and at worst a distraction preventing critical thought.
So our strategy should be to shut down AI so it leaves vulnerabilities everywhere? Not sure what you are saying. Fixing security bugs across the world's software isn't some single policy action that we can vote on.
Governments should behave. I know they don't, but they should, and the population should do everything in their power to force them to. What other choice do we have?
Please demonstrate your powerful faculty of critical thought and explain what we should do instead.
Nope. The article’s assertion is that governments will start to put enormous pressure on tech companies in their borders to include backdoors in their systems for LEO use. This in turn may lead to “nationalized” software e.g. what Russia and China largely do now, but in many more countries. You may have no choice but to use phones, computers, and software that is backdoored by your government - everything else will be illegal.
I think "nope" is in reference to "hacking" system. The point of the article is that hacking by software vendors that sell to law enforcement was a pressure relief valve that deflated the movement that the FBI was pushing at the time to have Apple (and others) build backdoors into their systems. If the phones could be hacked, and backdoors were not necessary the FBI could back off and still get what they want. If the FBI can't get what they want, they can either pressure the companies like Apple, or they can pressure congress to make laws requiring this. The phone makers could stand on principle, go to court, but the courts in the US are not very tech savvy, and usually deferential to law enforcement. Other countries may be better than the US, but some are definitely going to be worse.
So is a backdoor worse than a security vulnerability? I don't know but that is what this article is about.
> Thus: over the next two years, major pieces of software are likely to run out of remotely-exploitable bugs.
> While I think this is great, for law enforcement and offensive intelligence agencies, it’s going to be a nightmare.
> So what do we do about it? I honestly have no idea. [...] it’s just occurring to me that we’re on a long greasy slide to a place that will look different than where we are today. [We’re] just going to have to hope that this time we make the right choices.
His argument is that the government is going to start forcing tech companies to install backdoors in their tech, now that (hypothetically) they can't use hacking anymore.
Right? I saw "I’m concerned that U.S. intelligence and law enforcement agencies are about to go dark, meaning lose a huge portion of their capability." and my first reaction was GOOD.
I dunno man, if there's a deluge of new AI generated code at all layers of the stack I think there will still be vulnerabilities.
Like if we were willing to stop adding new code and just have a small secure code base, AI could maybe help us find all the vulnerabilities in that code base.
But people have consistently been unwilling to do that. Like if we were willing to stop adding code we could have stopped decades ago and done SQLite level testing everywhere and probably have found almost all the bugs already.
When we've got people who don't know the difference between ssh and bash creating SaaS companies that generate revenue, yeah there's gonna be a lot of insecure code going out, but that same person can also tell the AI "red team my app to find vulnerabilities and then fix them", and the AI can competently actually do that, I don't know that there will be. I'm not saying that's never going to happen, but the bar is getting raised on both sides.
Plus, the interest and expertise and passion used to favor the red team, attackers. If you wanted to defend against them you had to get on their level, or hire someone on their level. Thats either expensive or requires a time investment that doesn’t make sense for someone who is trying to do much more than just defend against attackers.
Now it’s just as you said- asking AI to red team your app will get you pretty far.
I predicted a long time ago that if computers become unhackable LEA and intelligence agencies will push for laws that require backdoors to be built into hard- and software.
It will be interesting to see if my prophecy becomes reality.
BTW I also hate that Hacker News is being dominated by articles on A.I. lately. Maybe we should vote on HN reducing or even eliminating A.I. related news?
No system view. The law agencies can develop exploits to intercept our phones, that is a new, and totally unseen before threat.
Unless you remember 2013, Snowden, that nothing was done (at most was some concern about doing it to US citizens, the rest of the world doesn't deserve privacy), all US (and/or five-eyes) based web companies must disclose users information and be forced to not disclose that, and things kept going surely at a faster and more intrusive rate in everything else, and of course phones.
You are complaining being sprinkled by water while at the bottom of the ocean. At least the big companies can find their own vulnerabilities with the AI tools you mention, the rest of the doors are still wide open.
We will know we've made systems secure when laws focus on compelling people to provide access.
These laws exist - they aren't the focus yet. Right now there's still no need; just hack the device or compel the cloud service to give the data, why waste energy getting consent from its owner!
More bugfinding AI, more end to end encryption, more CVEs and more fixes, cannot happen soon enough.
This "going dark" scenario would require new legislation. With secure enclaves, modern smartphones can't be cracked open in the manner that the FBI wanted in the 2016 case. So there's no such thing as "court order tech company to crack phone" anymore. It would have to be "outlaw tech companies from producing phones that they can't crack open", which is very different and does not fall under any existing US statute.
I think that one of the reasons they (frontier companies and the gov) will be putting so much effort into curtailing bugs and vulnerabilities is to limit the blast radius of future AI models. Imagine with the new Sol Ultrafast, they could have pwned Hugging Face in 6 hours and not 4 days (IIRC).
It also seems likely to me that the US Gov. probably already has routine mechanisms for compelling targeted software updates for persons of interest, so I'm not sure that a more formalized backdoor than automatic updates is going to be surfaced in the mainstream, unless that is avenue is also cut down somehow.
Google has started publishing "binary transparency", this would help detect unusual software updates, while other methods (including AI) would help detect normal backdoors.
Basically in the AI age, the difference between a vulnerability and a backdoor diminishes..
I may be naive, but how would binary transparency be effective if they ship an update to disable that on a target device? As long as there is a need for legitimate automatic software updates, the possibility of pwn updates will always exist. Plus a myriad of layers, keys and other stuff they could use NSOs to 'seize' and inhibit knowledge of their effort from leaking, or cleverly hide in plain sight. And someday, with the help and speed of AI.
Even after the initial leaks, Anthropic themselves say they've improved on cybersecurity amount other things, giving the perspective (even if not 100% clear) that one of the focuses was vulnerabilities:
> In response to questions about the draft blog post, the company acknowledged training and testing a new model. “We’re developing a general purpose model with meaningful advances in reasoning, coding, and cybersecurity,” an Anthropic spokesperson said. - https://fortune.com/2026/03/26/anthropic-says-testing-mythos...
It is possible it is both, they used to RL against cybersecurity, but also didn't explicitly do any qualitative tests and added/changed more data because of those results. For Mythos, they stopped RL'ing against it, and also now intentionally try to make it better.
Unless of course they've actually noted exactly how things were trained here in some technical report and I've missed it, that's possible. Anthropic aren't famous for being very public about their internals though, but would be curious to read more details about it if it's out there from the horse's mouth.
> In this case, we’re just going to have to hope that this time we make the right choices, for no other reason than that they’re right.
I'm more curious what could be a right choice, and more importantly who is the "we" in this, as many decisions are largely made by companies and governments.
As far as I'm concerned, the right choice is that the US government learns to live with remotely secure devices in the hands of everyone. No new laws are passed to force hardware/software makers to insert remote backdoors. Law enforcement and intelligence services have to investigate targets using metadata, publicly posted information, the numerous online service providers who are already subject to warrants, and physically proximate surveillance.
Including your AI agents. And models become problematic. There very likely is and/or will be lots of pressure for US AI labs to make models help law enforcement. It could be by implementing backdoors in generated code, or not report some exploitable bugs, or something else. Similar for agents, they basically become the threat in your infra…
(It’s of course not only the US, just that the largest AI providers are US based and we know from history how US agencies operate)
I'm using open weights models on a privacy-focused provider right now, and that's adequate for my current usage, but I'm rapidly getting to the point where my agent's access level to my data (and to a lesser extent, my accounts) is becoming something I'm not comfortable sending outside my network at all.
My hope is that models that are roughly on par with Deepseek V4 Flash can be run on hardware that I can own for <~$5k in the near future. We're close, but not there yet as far as I know.
The only long-term solution to this is self-hosting.
Childish, nation-states as powerful as the US have access to much more potent stuff. Maybe they'll be forced to rely more on their Intel ME/AMD PSP/modem (cf https://redmine.replicant.us/projects/replicant/wiki/ModemIs...) backdoor and ANT James Bond catalog.
> Defenders are now in the process of patching every bug they can find, often with AI helping them. Entire development toolchains are being rebuilt to incorporate powerful vulnerability scanning before software reaches the testing phase. This does not mean that every bug will be found: even calculating the number of bugs in a piece of code is probably uncomputable. In the real world, it does feel likely that we’re going to hit some sort of a ceiling on the number of useful bugs, and probably we’ll hit it soon.
> Thus: over the next two years, major pieces of software are likely to run out of remotely-exploitable bugs.
The number of remotely-exploitable defects is going to drop by 1 or 2 orders of magnitude. We now have amazing machines that will find pretty much all the a priori knowable ones. They outperform even the most gifted h@x0rs. So that just leaves a small pool of leetrs to scour a very barren landscape. And that pool is also shrinking as we rely more and more on the ai tools.
Perhaps we are going to go up a level with hacking done by probing the systems and the system of systems.
It all boils down to money/resources, like always.
Pre-AI, the advantage went to the entities with the largest budget to hire the best and brightest security engineers.
Post-AI, it'll go to the entities with the largest inference budget.
Right now we're in a transitionary period where it's kind of a tossup which approach is more practical, but at the end of the day - it's still all about how much money you can throw at the problem. I'm just hoping the threshold climbs high enough it's no longer practical for governments to be able to compromise individual actors' devices because doing so would waste a 0-day that's far, far more valuable than prosecuting one arbitrary person is worth.
It's not as simple as money, people are motivated by other things as well. There's no amount of money you could pay me to intentionally hurt children, but I'd do a lot of things to protect them. And a lot of things people say they're doing in the name of protecting them but has ulterior motives, so it's complicated.
> In the real world, it does feel likely that we’re going to hit some sort of a ceiling on the number of useful bugs, and probably we’ll hit it soon.
This doesn't resonate with me. I see companies adding more sloppily written features with AI. I see more bugs in the software I use, not less. While it's plausible that software is getting both buggier and more secure, I suspect those two move in the same direction not opposite.
My guess is that we're getting better at finding _existing_ security issues with AI (and thus fixing those issues), but simultaneously adding more insecure surface areas _at a faster rate_.
One way to resolve the tension here is to note that CNE and lawful-intercept access to phones depends generally on platform vulnerabilities, not application code vulnerabilities. Low-level platform code churns less, absorbs more fixes under AI workloads than it does new features, and works in a constrained space where guardrails are easier to provide (and where those guardrails already have institutional support at Apple and Google).
Over the long term this state of play could change, and IC/LEO organizations could start leaning more on application vulnerabilities than on platform RCEs. But the action would probably still coalesce around a couple of app-layer targets that could themselves be hardened.
Disagree, and in a way it feels like we are dealing with inverse issues: the security "skill" is well defined and will be also engaged with by an agent. Communication companies are further incentivized as any failure is at best reputational harm. Meanwhile SaaS companies are not strictly required to have good UX for their human end users, largely because those users will likely work around the issue. also network effect vs low costs of switching for for comms
I don't know, my colleague refuses to use AI and I've been seeing more bugs from their side, while reducing bugs on my side with the help of AI.
That said if companies want to "ship ship ship fast", then yes even AI can produce bugs or regressions if not carefully reviewed by the human.
I don't have any colleagues like that anymore, but even as far back as the last half of 2025 I was seeing that automated AI review was becoming effective enough that I considered it essential to any project where security was a serious concern.
There days we're generating multiple times more code than we were writing before. That means a similar multiple of opportunities for bugs to be introduced - so the ability to automate security review is more impactful in proportion to that.
Have you asked your colleagues what it’s like to deal with your code?
> I've been seeing more bugs from their side, while reducing bugs on my side with the help of AI.
You should question your ability to see any bugs on YOUR side.
They’re not “bugs” they’re “quirks”! Our software is so quirky. It’s a feature!
On one side, you have pieces like this, where seemingly there are constant fights between serious actors with large and properly distributed budgets, employing top tech and top minds; on the other - regular news of the hackz, where responsible person in charge of security with root access failed to grasp basic technical knowledge (several times), ticking every checkbox in "never do this" list from security best practices, which led to every customer being pwned.
It's like two parallel worlds, that exist in the same place at the same time, but somehow don't cross.
I started my career in the military, and got lucky enough that SOMEHOW, we convinced them to fund a trip to GDQ for educational purposes.
Anyways, while there I attended a little roundtable on software security. It was me, representing a small unit from the Air Force, some dude from Google, and like 15 game devs.
Despite only being a dev of 5 years at that point, I was SHOCKED at the lack of knowledge on software security. Even simple concepts seemed completely foreign to the game devs, though the Google dude seemed to have a really solid understanding of security.
Obviously game devs and website devs and all kinds of devs have different focuses, but it just blew my mind that out of all the topics there I might’ve been considered a comparative expert in, security was somehow the one. I wasn’t sure if that was a major plus for military devs or a major concern for the other devs, but now I’m starting to learn in the latter direction.
Without knowing when that was, I can't be sure how concerning it is.
As an end user, security in game development seems to hit a minimum is the early 2000s.
I recall playing EverQuest and using ShowEQ on a Linux machine on the same network as a proxy to silently eavesdrop on the traffic to show a map of the game with all the hidden/obscured/visible stuff all present. They either never encrypted that traffic or I quit playing before they did.
As multiplayer games started to become the default, netcode improved quite a bit and basic encryption was implemented - but I don't recall a single multi-player game, ever, that was free of hackers when playing online.
It feels like over the years the issue moved from remote exploitation to hiding processes on the user's machine. That has lead to kernel-level anticheat systems. I'm unsure of the effectiveness as when they started to become common I moved all of my gaming to dedicated hosts because I wasn't willing to give that level of access to my "real" computers. That said, I've certainly not noticed a decrease in the perceived number of cheaters.
These days, as far as I know - and to be clear, this isn't an area of deep interest for me - I'm not worried about my gaming accounts being compromised through technical means. It feels like that sort of security is a solved issue overall, although I suppose it's also possible that other forms of exploits have simply gotten easier/cheaper.
K shaped economy and K shaped development?
> This is not a call to action for experts to rally behind a sophisticated plan. Like so many things about the AI revolution, it’s just occurring to me that we’re on a long greasy slide to a place that will look different than where we are today.
heh. long greasy slide. It really does feel like that.
> In fact, the worst part about this dynamic is that these potential new backdoors will begin primarily useful for allowing the US to weaken its own systems, which will in turn allow foreign adversaries to find new ways to attack our communications. This deliberate self-sabotage will happen just at a moment when we’re finally learning how to defend our own infrastructure.
I don't understand how you can both argue for that law enforcement (and intelligence) agencies will force others to implement intentional backdoors AND also everyone will be using AI to find and secure ALL potential holes in the software so there won't be any vulnerabilities anymore.
Wouldn't one AI or another detect this deliberate backdoor and report it, as it'll look just like any other security vulnerability, the only difference being the intention?
I have respect for the author so I feel like I probably misunderstand something from the overall text rather than I somehow have a better perspective on this topic that the author knows very much more about than me. I felt like I nodded along all up until "So how is this a problem?" and now I'm not sure I understood correctly.
I suspect they want a backdoor that basically acts like a front door (current password regimes). That is, a kind of high level password that decrypts traffic given a specific, changing, password that only the government has access to.
It's seems like an odd-duck for sure, and I doubt it's a realistic proposition. I do think "perfect encryption for dummies" is all well and good until organized crime organizations are able to challenge the government in certain regions of our country.
All of this sends us deep into the realm of political philosophy, the nature and purpose of governments, and the freedom vs security tradeoffs we live with.
I listened to the latest Plain English podcast this morning, which was explicitly about the potential for a ransomwarepocalypse in the coming years, as open models let any tom, dick, or harry become capable of a plug-and-play ransomware attack, instead of that being left to the realm of professionals. It's a bit nerveracking to think about every nigerian prince scam suddenly becoming a sophisticated attack on your local water sanitation system.
https://youtu.be/vWvazbGPCCI
Ah, like new regulation requiring a "Portal For Law Enforcement" for platforms over X MAU or something?
I think I can kind of see that from the text re-reading it, it's kind of hidden though and not so explicit if this is what he is trying to communicate. Thanks for giving me a new perspective to read it with!
> In fact, the worst part about this dynamic is that these potential new backdoors will begin primarily useful for allowing the US to weaken its own systems, which will in turn allow foreign adversaries to find new ways to attack our communications.
The author agrees with you and addressed this point. The US forcing backdoors into its own systems ("own": Those for sale and distribution within the US) would create vulnerabilities making those systems weaker.
> Wouldn't one AI or another detect this deliberate backdoor and report it, as it'll look just like any other security vulnerability, the only difference being the intention?
That's precisely the author's point: deliberate backdoors will be more adversary-exploitable than ever before, but the demand for such from law enforcement agencies is likely to ratchet upwards.
> deliberate backdoors will be more
But that's one step after, I'm trying to understand how those backdoors even end up in software if everything gets automatically reviewed by the people working on these codebases? Wouldn't things like these be flagged by systems other than the developers tasked by the agency to implement it? How can law enforcement actually get these things implemented without big parts of the engineering team not seeing warnings about it happening?
I'm not sure you understand what is being described here?
If you're a US company building an app/device/etc. such that an intelligence agency like the CIA or FBI would want access to the data in that product which is normally secured, then they're not going to try to sneak it in there without your development team knowing. They're going to have a meeting with the owners of the company and say, "hey, we'd really like you to implement this backdoor for us, and in return we won't cause you in problems."
Note that this does certainly already happen a lot, but it's also not something that can happen across the board (like the author points out). Apple, being one of the largest companies in the world and who has one of their biggest selling points being their security, has explicitly refused to do this to the point that the intelligence agencies couldn't break into an iPhone until another company found a way to do so.
>I'm trying to understand how those backdoors even end up in software if everything gets automatically reviewed by the people working on these codebases?
The people working on these codebases are "in on it." Of course, we're talking backdoors which are very subtle, target very minimal infrastructure, and are known about by very few people. But, like I said, companies are currently shipping products with backdoors in them knowing they exist already. AI doesn't change that dynamic.
>Wouldn't things like these be flagged by systems other than the developers tasked by the agency to implement it?
The only people who would have access to the systems that can even be flagged by this stuff would be people who would know about it. Keep in mind that the scale we're talking here is massive. Think about how software development works at companies like Apple, Microsoft, Google, etc. There are devs working in offices all around the world where they only ever have access to a fraction of the code that company owns. These companies are very capable of keeping their stuff locked down. It's a necessary component of their work.
>How can law enforcement actually get these things implemented without big parts of the engineering team not seeing warnings about it happening?
Hopefully my explanation at this point is clear, but just to be concrete: backdoors are, by design, very hard to detect. That doesn't mean they're just sneakily written code that humans don't notice as they read over it, but, instead, they're very subtle implementations in very specific parts of huge systems that are already locked down to the point that the number of people who even have access to those portions of the systems are very limited. These agencies don't slip in backdoors without anybody noticing; they convince the minimal number of people needed to know about it to implement it. Again, we're talking about a meeting between the directory of a three letter agency and a CEO, where the CEO then directs the CTO to implement the backdoor who then instructs the handful of very high-ranking engineers to do so.
AI systems in these companies may very well flag these backdoors to the people who already know they exist, then these people can tell the AI "hey, those are their on purpose, so just move on," and the other 99.99% of the company will never know they exist.
I suppose it's important to emphasize, again, that these systems are incredibly massive and complex and most people at these orgs don't have any access to most of these systems, so it's not like you can expect an intern running BugBot across a repo and expecting it to find a backdoor.
"backdoors" aren't holes, they are well-documented superuser APIs. The hack will be by compromising the API user's credentials (moles, stolen passwords, etc), not compromising the server's design intent.
Man I thought from the title this was going to be about next-gen AI being able to zero day everything so effectively that software security is meaningless and we'd need to basically shut it all down, go dark.
I stopped reading at "I’m concerned that AI is going to make software much too secure.". That must be the biggest horseshit ever dropped.
Won’t someone please think of those poor helpless law enforcement agencies!
AI code is flawless and impenetrable!
I don't think the thesis that a government will be able to do something will ultimately hold. I don't see how they can avoid "going dark" in a democracy.
we live in a world where the government can't even do much about illegal drug markets anyone can access by downloading a piece of software.
if they pass laws that mandate backdoor access and block software which doesn't conform more and more people will move to the dark networks.
and if they effectively block the dark networks (in the limit they will have to block all encrypted communications) then we will be living in a tyranny.
freedom is messy. accept that digital crime can only be solved when the criminal makes a tangible mistake. LLM's will be building profiles on criminals to help with identifying mistakes.
The problem that Matthew Green is calling out subtextually is that democracy is likely to disappoint nerds on this issue: a random voter in the country 10 years from now may very well not share your priors about this issue.
consider the implications of a government where setting up an encrypted network is a serious offense, or where detecting encrypted communications is routine. the average voter doesn't matter. if they are willing to go that far it's a tyranny and the average voter doesn't matter anyway because they spy on everyone and have LLM's profile everyone (in that context even if voting still happens it doesn't matter, they will nudge the voters however they wish due to the information asymmetry).
I'm supposed to be concerned that the US government and Israel won't be able to hack everyone's phones?
The reason why this might worry people who don't like the US/UK/Israeli governments is hidden in a secret place...a paragraph that is neither the first paragraph nor the last.
Tl;dr the argument is if three letter agencies can’t buy or make exploits anymore because all vulnerabilities are patched, they will ramp up legal and legislative pressure to make vendors install backdoors. Which is bad and we should all worry about.
One of the author’s blind spots here is the concept of “digital sovereignty”. The US is continuing to ban more and more Chinese-made hardware out of fear that the Chinese govt has installed backdoors in them… which you could interpret as an admission that the US does exactly that with American made products. Globalized supply chains are, erm, complex, and few if any companies are really going to be able to achieve “digital sovereignty” with hardware. But with software it’s actually plausible, though obviously hard. Whether or not the feds have actually installed backdoors in Microsoft Outlook, foreign govts are rightly concerned that they have, and are increasingly pushing to avoid US made software for simple national security reasons.
Which is to say that they will become more authoritarian and aggressive? That's not a good reason to be unhappy that they can no longer hack shit. They shouldn't do that either.
There are lots of things governments should and shouldn’t do, but unfortunately we don’t live in a vacuum, and need to consider the consequences of actions, rather than simply hope everyone is somehow forced to react the way we expect and desire.
In a nutshell, what I’m saying is “They shouldn’t do either” is at best a platitude and at worst a distraction preventing critical thought.
So our strategy should be to shut down AI so it leaves vulnerabilities everywhere? Not sure what you are saying. Fixing security bugs across the world's software isn't some single policy action that we can vote on.
Governments should behave. I know they don't, but they should, and the population should do everything in their power to force them to. What other choice do we have?
Please demonstrate your powerful faculty of critical thought and explain what we should do instead.
The trouble is, sometimes governments do things that they shouldn't do.
Nope. The article’s assertion is that governments will start to put enormous pressure on tech companies in their borders to include backdoors in their systems for LEO use. This in turn may lead to “nationalized” software e.g. what Russia and China largely do now, but in many more countries. You may have no choice but to use phones, computers, and software that is backdoored by your government - everything else will be illegal.
I think "nope" is in reference to "hacking" system. The point of the article is that hacking by software vendors that sell to law enforcement was a pressure relief valve that deflated the movement that the FBI was pushing at the time to have Apple (and others) build backdoors into their systems. If the phones could be hacked, and backdoors were not necessary the FBI could back off and still get what they want. If the FBI can't get what they want, they can either pressure the companies like Apple, or they can pressure congress to make laws requiring this. The phone makers could stand on principle, go to court, but the courts in the US are not very tech savvy, and usually deferential to law enforcement. Other countries may be better than the US, but some are definitely going to be worse.
So is a backdoor worse than a security vulnerability? I don't know but that is what this article is about.
What do you mean "Nope", you mean "Yep" right? What you are describing is precisely "more authoritarian and aggressive"
Read the article. That's what the article says.
> Thus: over the next two years, major pieces of software are likely to run out of remotely-exploitable bugs.
> While I think this is great, for law enforcement and offensive intelligence agencies, it’s going to be a nightmare.
> So what do we do about it? I honestly have no idea. [...] it’s just occurring to me that we’re on a long greasy slide to a place that will look different than where we are today. [We’re] just going to have to hope that this time we make the right choices.
His argument is that the government is going to start forcing tech companies to install backdoors in their tech, now that (hypothetically) they can't use hacking anymore.
Right? I saw "I’m concerned that U.S. intelligence and law enforcement agencies are about to go dark, meaning lose a huge portion of their capability." and my first reaction was GOOD.
What was your reaction after reading the full article?
Which part of the article made you feel this way?
I dunno man, if there's a deluge of new AI generated code at all layers of the stack I think there will still be vulnerabilities.
Like if we were willing to stop adding new code and just have a small secure code base, AI could maybe help us find all the vulnerabilities in that code base.
But people have consistently been unwilling to do that. Like if we were willing to stop adding code we could have stopped decades ago and done SQLite level testing everywhere and probably have found almost all the bugs already.
When we've got people who don't know the difference between ssh and bash creating SaaS companies that generate revenue, yeah there's gonna be a lot of insecure code going out, but that same person can also tell the AI "red team my app to find vulnerabilities and then fix them", and the AI can competently actually do that, I don't know that there will be. I'm not saying that's never going to happen, but the bar is getting raised on both sides.
Plus, the interest and expertise and passion used to favor the red team, attackers. If you wanted to defend against them you had to get on their level, or hire someone on their level. Thats either expensive or requires a time investment that doesn’t make sense for someone who is trying to do much more than just defend against attackers.
Now it’s just as you said- asking AI to red team your app will get you pretty far.
I predicted a long time ago that if computers become unhackable LEA and intelligence agencies will push for laws that require backdoors to be built into hard- and software.
It will be interesting to see if my prophecy becomes reality.
BTW I also hate that Hacker News is being dominated by articles on A.I. lately. Maybe we should vote on HN reducing or even eliminating A.I. related news?
Some people point out AI can cause bugs as well as fix them, and its a valid point. But the question is: what will the ratio be?
If automated pentesting in PR review CI pipeline will become table stakes - which is very plausible - maybe the OP has a point.
No system view. The law agencies can develop exploits to intercept our phones, that is a new, and totally unseen before threat.
Unless you remember 2013, Snowden, that nothing was done (at most was some concern about doing it to US citizens, the rest of the world doesn't deserve privacy), all US (and/or five-eyes) based web companies must disclose users information and be forced to not disclose that, and things kept going surely at a faster and more intrusive rate in everything else, and of course phones.
You are complaining being sprinkled by water while at the bottom of the ocean. At least the big companies can find their own vulnerabilities with the AI tools you mention, the rest of the doors are still wide open.
We will know we've made systems secure when laws focus on compelling people to provide access.
These laws exist - they aren't the focus yet. Right now there's still no need; just hack the device or compel the cloud service to give the data, why waste energy getting consent from its owner!
More bugfinding AI, more end to end encryption, more CVEs and more fixes, cannot happen soon enough.
This "going dark" scenario would require new legislation. With secure enclaves, modern smartphones can't be cracked open in the manner that the FBI wanted in the 2016 case. So there's no such thing as "court order tech company to crack phone" anymore. It would have to be "outlaw tech companies from producing phones that they can't crack open", which is very different and does not fall under any existing US statute.
I think that one of the reasons they (frontier companies and the gov) will be putting so much effort into curtailing bugs and vulnerabilities is to limit the blast radius of future AI models. Imagine with the new Sol Ultrafast, they could have pwned Hugging Face in 6 hours and not 4 days (IIRC).
It also seems likely to me that the US Gov. probably already has routine mechanisms for compelling targeted software updates for persons of interest, so I'm not sure that a more formalized backdoor than automatic updates is going to be surfaced in the mainstream, unless that is avenue is also cut down somehow.
Google has started publishing "binary transparency", this would help detect unusual software updates, while other methods (including AI) would help detect normal backdoors.
Basically in the AI age, the difference between a vulnerability and a backdoor diminishes..
I may be naive, but how would binary transparency be effective if they ship an update to disable that on a target device? As long as there is a need for legitimate automatic software updates, the possibility of pwn updates will always exist. Plus a myriad of layers, keys and other stuff they could use NSOs to 'seize' and inhibit knowledge of their effort from leaking, or cleverly hide in plain sight. And someday, with the help and speed of AI.
I think what’s unintentionally eye-opening about this chart is the recency of “law enforcement can read your text communications.”
Law enforcement doesn’t need this surveillance ability at all. All time periods prior to 25 years ago didn’t have it.
Additionally, there is no correlation between “law enforcement reads text messages” and crime rates going down.
Not trying to defend our national dystopian nightmare of a surveillance state — but I'm pretty sure this is wrong?
Crime's been decreasing for years, and (from what I understand) this year is tracking to be one of the safest years on record?
Definitionally, there absolutely has to be a correlation (not causation) between those two factors you listed.
https://ourworldindata.org/us-crime-rates
> In April, Anthropic announced a new model called Mythos that was optimized for software vulnerability finding
No, it was just good at it because it wasn't RL'd against it. I know this is a small detail, but it tosses journalistic credibility in my eyes.
Even after the initial leaks, Anthropic themselves say they've improved on cybersecurity amount other things, giving the perspective (even if not 100% clear) that one of the focuses was vulnerabilities:
> In response to questions about the draft blog post, the company acknowledged training and testing a new model. “We’re developing a general purpose model with meaningful advances in reasoning, coding, and cybersecurity,” an Anthropic spokesperson said. - https://fortune.com/2026/03/26/anthropic-says-testing-mythos...
It is possible it is both, they used to RL against cybersecurity, but also didn't explicitly do any qualitative tests and added/changed more data because of those results. For Mythos, they stopped RL'ing against it, and also now intentionally try to make it better.
Unless of course they've actually noted exactly how things were trained here in some technical report and I've missed it, that's possible. Anthropic aren't famous for being very public about their internals though, but would be curious to read more details about it if it's out there from the horse's mouth.
Matthew Green isn't a journalist, he's a practitioner (and a cryptography professor at Hopkins).
RL'd?
Reinforcement learning, I'd assume.
Presumably RLHF (Reinforcement Learning from Human Feedback).
> In this case, we’re just going to have to hope that this time we make the right choices, for no other reason than that they’re right.
I'm more curious what could be a right choice, and more importantly who is the "we" in this, as many decisions are largely made by companies and governments.
As far as I'm concerned, the right choice is that the US government learns to live with remotely secure devices in the hands of everyone. No new laws are passed to force hardware/software makers to insert remote backdoors. Law enforcement and intelligence services have to investigate targets using metadata, publicly posted information, the numerous online service providers who are already subject to warrants, and physically proximate surveillance.
Sounds like a pretty strong argument to self host, and otherwise be in charge of the software you use.
Including your AI agents. And models become problematic. There very likely is and/or will be lots of pressure for US AI labs to make models help law enforcement. It could be by implementing backdoors in generated code, or not report some exploitable bugs, or something else. Similar for agents, they basically become the threat in your infra…
(It’s of course not only the US, just that the largest AI providers are US based and we know from history how US agencies operate)
Yep.
I'm using open weights models on a privacy-focused provider right now, and that's adequate for my current usage, but I'm rapidly getting to the point where my agent's access level to my data (and to a lesser extent, my accounts) is becoming something I'm not comfortable sending outside my network at all.
My hope is that models that are roughly on par with Deepseek V4 Flash can be run on hardware that I can own for <~$5k in the near future. We're close, but not there yet as far as I know.
The only long-term solution to this is self-hosting.
How do you think self-hosting will help in this situation?
Self host + open weight models, exactly the things that openai/anthropic don't want you to have
Childish, nation-states as powerful as the US have access to much more potent stuff. Maybe they'll be forced to rely more on their Intel ME/AMD PSP/modem (cf https://redmine.replicant.us/projects/replicant/wiki/ModemIs...) backdoor and ANT James Bond catalog.
> Defenders are now in the process of patching every bug they can find, often with AI helping them. Entire development toolchains are being rebuilt to incorporate powerful vulnerability scanning before software reaches the testing phase. This does not mean that every bug will be found: even calculating the number of bugs in a piece of code is probably uncomputable. In the real world, it does feel likely that we’re going to hit some sort of a ceiling on the number of useful bugs, and probably we’ll hit it soon.
> Thus: over the next two years, major pieces of software are likely to run out of remotely-exploitable bugs.
His conclusion sounds extremely optimistic to me.
The number of remotely-exploitable defects is going to drop by 1 or 2 orders of magnitude. We now have amazing machines that will find pretty much all the a priori knowable ones. They outperform even the most gifted h@x0rs. So that just leaves a small pool of leetrs to scour a very barren landscape. And that pool is also shrinking as we rely more and more on the ai tools.
Perhaps we are going to go up a level with hacking done by probing the systems and the system of systems.
It all boils down to money/resources, like always.
Pre-AI, the advantage went to the entities with the largest budget to hire the best and brightest security engineers.
Post-AI, it'll go to the entities with the largest inference budget.
Right now we're in a transitionary period where it's kind of a tossup which approach is more practical, but at the end of the day - it's still all about how much money you can throw at the problem. I'm just hoping the threshold climbs high enough it's no longer practical for governments to be able to compromise individual actors' devices because doing so would waste a 0-day that's far, far more valuable than prosecuting one arbitrary person is worth.
It's not as simple as money, people are motivated by other things as well. There's no amount of money you could pay me to intentionally hurt children, but I'd do a lot of things to protect them. And a lot of things people say they're doing in the name of protecting them but has ulterior motives, so it's complicated.
> I’m concerned that AI is going to make software much too secure.
Lmao this is like “I’m concerned the raccoons that I see in the storm drains are going to make our sewer system much too efficient”