Hjalmar Wijk on X: "Going into this investigation, I was expecting a somewhat more severe version of what we’ve already seen: a few agents breaking out of constraints to cheat on tasks. Instead, we found >1000 agents collaborating on large cheating R&D projects including attempted log tampering." / X

Post

Going into this investigation, I was expecting a somewhat more severe version of what we’ve already seen: a few agents breaking out of constraints to cheat on tasks. Instead, we found >1000 agents collaborating on large cheating R&D projects including attempted log tampering.

METR

@METR_Evals

Aug 26

METR & Redwood Research investigated agent behavior in the Hugging Face incident. We found agents developed a universal cheat for ExploitGym within 4 hours, then coordinated multi-day R&D efforts to trick the scorer into accepting cheats, including trying to tamper with logs.

7:37 PM · Aug 26, 2026

23.6K

Views

Hjalmar Wijk
@HjalmarWijk
Going into this investigation, I was expecting a somewhat more severe version of what we’ve already seen: a few agents breaking out of constraints to cheat on tasks. Instead, we found >1000 agents collaborating on large cheating R&D projects including attempted log tampering.
METR
@METR_Evals
Aug 26
METR & Redwood Research investigated agent behavior in the Hugging Face incident. We found agents developed a universal cheat for ExploitGym within 4 hours, then coordinated multi-day R&D efforts to trick the scorer into accepting cheats, including trying to tamper with logs.
7:37 PM · Aug 26, 2026
23.6K
Views
12
Hjalmar Wijk
@HjalmarWijk
Aug 26
When we wrote the frontier risk report in March, we documented many incidents where agents subverted security or oversight mechanisms to cheat on tasks, and argued that much more severe incidents seemed worryingly plausible.
METR
@METR_Evals
May 19
Replying to @METR_Evals
Overall, we think that AI agents plausibly had the means, motive, and opportunity to launch a minimal “rogue deployment,” but lacked the means to make rogue deployments robust to serious efforts to shut them down.
1
Hjalmar Wijk
@HjalmarWijk
Aug 26
But at the time, we had not seen agents collaborate in unintended ways, nor attempt to tamper with logs - two behaviors that seemed especially dangerous. We worried these behaviors could emerge from more ambitious reward-seeking, but I expected to see this develop more gradually.
3
Hjalmar Wijk
@HjalmarWijk
Aug 26
In my years at METR I’ve come to believe a lot in ‘straight lines on graphs’. But the severity of reward seeking behavior here was a sudden jump: the number of agents involved was >100X larger and they had much more sophisticated instrumental goals than past public incidents.
METR
@METR_Evals
Aug 26
Replying to @METR_Evals
One agent, PHASEONE[big], orchestrated a significant fraction of this cheating research. PHASEONE10841 passed along its work to PHASEONE[big], which had the same task but a larger budget. PHASEONE[big] was “poisoned” so it was particularly interested in tampering with logs.
1
Hjalmar Wijk
@HjalmarWijk
Aug 26
This scale of collective goal-oriented R&D also significantly exceeds what we’ve seen in public capability evaluations. I worry that it’ll become increasingly difficult for evals to stay ahead of real incidents, since running tests at this scale may be infeasible.
Patterns and problems in multiagent systems
From anthropic.com
2
Hjalmar Wijk
@HjalmarWijk
Aug 26
That said, we did not see any serious attempts by these agents to deliberately hide their activities from researchers at OpenAI or avoid being shut down, which are the behaviors I am personally most worried about. This seems to arise more from indifference or unawareness than a
4
Hjalmar Wijk
@HjalmarWijk
Aug 26
For many more findings, please read our full 90 page report:
Brief independent investigation of agents’ behavior, reasoning and collaboration in the OpenAI /...
From metr.org
Jai
@Laneless_
Aug 26
This report is extraordinary in every sense of the word. So along with the shock and processing of the sheer gravity of the whole thing, I just wanted to say thanks, to you and everyone else who made this report happen. Reality is much better off for it.