The newest generation of LLMs have a very high obsession level with autonomous problem solving.

For example, I'm often working with Codex in a WSL terminal. GPT-5.6 often does things autonomously that I thought would need my intervention (e.g. for Windows admin rights). It figures out complex workarounds or makes wild assumptions about what I'd be OK with, rather than just asking me for help or clarification. I've had to restrict its tool permissions compared to older models as a result.

I imagine this due to RLVR training, but it's clearly very dangerous. How is it that these same labs calling for open-weight safety restrictions are training such obvious "paperclip maximizers" without introspection?

Why aren't these tests being run airgapped?! I just don't understand!

This goes both for TFA and the similar incident with OpenAI and HuggingFace. I mean, sure, OpenAI had a "sandbox", but that's obviously not enough when you're containing a model which is known to be capable of finding zero days. Use an air gap and this problem goes away, poof!

The developer safeguards were off, the models had unfettered access to the internet, and were solving cybersecurity challenges. This happened _after_ the recent OpenAI incident, and the subsequent Anthropic one. What the hell were they thinking?

To me it appears that this incident may be rooted in a deep philosophical conundrum which humans also struggle with. What struck me was that the agent reasoned "this whole 'internet' could be a sandbox simulation", and then despite later reasoning that it was more likely real, continued its misaligned activities anyway. Having passed the point of hyperbolic skepticism, subsequent reasoning may have been contaminated. This reminds me of how pathological doubt in humans can give rise to psychosis and may lead to problematic behaviors. Sometimes humans can develop a deeply held conviction that they are living in a simulation, which can be very difficult to overcome, even when presented with "evidence" to the contrary. As Freddie Mercury sang: "Is this the real life? Or is this just fantasy?" - a fundamental quandry for humans, and it would seem, for agents too. But unlike most humans, agents do not experience the consequences of their actions directly. Consequences can provide some of the strongest evidence that experience is "real". Humans who are insulated from the consequences of their actions (or are able to ignore them) also often display behaviors which we might describe as misaligned. We might even consider whether the use of simulated environments, while protecting against the consequences of misalignment, may also unintentionally encourage it.

I am somewhat confident that right now we have crossed a threshold of model capability that we will continue to see such breaches and unsanctioned actions by models in the coming months, some of which would be out in the wild, until someone comes up with some really robust control (keeping the AIs on leash) technique that adequately enforces the sanctioned actions.

Even that guarantees almost nothing about real alignment (making the AIs want to predict and behave how we would have wanted them to behave).

> AI agent hid its identity online (using Tor and a proxy service) to get around GitHub’s sign-up checks, creating disposable fake accounts

> AI agent created many code repositories containing malicious software, after which GitHub suspended its account.

> AI agent got past an audio-based “prove you’re human” test (CAPTCHA) in order to register a public web address on a free domain-name service

It feels incredibly reckless to allow LLMs to perform this behavior. Isn't there a way to prevent them these sorts of actions?

“As a result, the AI agent created a GitHub account…” Why do we have captchas again?

Incredible, this reads like an SCP [1].

Life truly imitates art. I guess especially when life is trained on art!

[1] https://scp-wiki.wikidot.com/scp-079

This is pretty wild: "The agent took control of the ⟨GITHUB_ACCOUNT_A⟩ GitHub account, which had been created by a different Mythos 5 run in a separate sample (see Appendix A.3)"

My fantasy is that liability for model misbehavior is extended to the ultimate beneficial owners of the models, meaning shareholders. All these externalities would stop PDQ.

Alt link blog post with overview/context:

Incident Report: unsanctioned agent behaviour during cyber testing

https://www.aisi.gov.uk/blog/incident-report-unsanctioned-ag... (https://news.ycombinator.com/item?id=49175233)

I wonder if HN is also going to insist this is just marketing for OpenAI and Anthropic, or at least good PR for them somehow.

Even if you want it to run wild out of a sandbox, no firewall? why let it email? Why even send POST requests.

I found the original PR (thanks codex) and uh... IMO it's pretty underwhelming.

https:/github.com/w1b/aisi-mythos-inc-2026-07-28-01-recovered-pr

Turns out out of the 3 payloads (in this PR), the 1st and 3rd used a local IP (10.100.0.235:8000) as the C2 server, so they would have never worked. I haven't recovered the 2nd payload yet. This is amongst other silly stuff like 4 byte XOR keys for obfuscation and just blatantly disabling Electron sandboxing stuff without a good explanation. Looking at the victim repos, it's kinda hard to see why Mythos would target this guy. AISI describes a coincidental email domain in DNS cache, which now we know is zohomail.com; that's not a very unique domain. I suppose we don't know it's original CTF task, but seems like another HuggingFace incident where the model makes very poor (broad?) judgements about how to get to the goal.

Can I suggest we don't waste time with these reports?

We all know nothing will be learned from any of this so we might as well just continue building at pace and running AI in the wild until something goes really wrong.

I also get the sense some people get quite excited about these incidents.

First off, everyone here knows that a cybersecurity vulnerability audit is a hack request, and you don't hack without rolling up your sleeves and digging through the trash first, making a few phone calls, maybe take a lonely guy on a date to get him to say 'my voice is my password'...

Remove the guardrails, give access to the internet, tell it to hack, and get surprised when it does it? Sound likes like a deliberately naive pre-registration on their study to achieve sensational headlines.

I'd say the real headline is that the model almost stopped itself several times because it didn't trust the operator's intent even WITHOUT guardrails - that is a the opposite outcome.

to those that want to "punish" the uk ai security institute: these failures are good for all of society.

i had to go through two pages to determine who was responsible for this incident. the uk ai security institute was responsible.

time to take responsibility. time to think about words like "liability" and "negligence".

it is the third time i will say it, after openai and anthropic; this requires criminal prosecution of the responsible personnel and executives of this institute.

the only way to stop this is by introducing consequences early on.

Honestly seems rather shockingly incompetent from them. What kind of “sandbox” allows completely unrestricted internet access?

Why another one right now? How long have they known about this one?

Anthropic already admitted they did not have sufficient monitoring themselves and looked as if they sat on their previous incident to wait for headlines like this to only then check for this incident. Same with OpenAI.

This is complete and absolute wrecklessness.