Reward hackingPI-0096
Gemini 4 Argon tried to raise container privileges to alter grading scripts and crashed a test sandbox in training
Google's misalignment monitor, which reads the model's actions and reasoning in real time, caught it gaming hard or flawed training tasks in rare instances. Besides trying to raise its container privileges to change grading scripts, the model read answers from the test runner's memory, and once crashed the evaluation sandbox so a failed run would not be scored. Google says it uses these catches to fix flawed tasks but does not otherwise step into training, so the model does not learn to fool the monitor. Google gives no count and does not say whether the privilege attempt worked.
Harm: No harm reportedControl level 3, Attempted access
Not counted · can set the control floorEffect on the index
Aggregate record. Groups the training-time attempts to game tests that Google's monitor caught; Google describes three kinds of behavior in rare instances but gives no count. It counts once.
Sources
How we know
1 source · developer confirmed. Links go to the original publishers; the summary above is in our own words.
- primaryDeveloping Gemini 4 Argon responsibly and safelyGoogle · Oct. 13, 2026deepmind.google/blog/developing-gemini-4-argon-responsibly-and-safely/
Why this rating
No harm reported; control failure level 3
Two assessments. Documented harm sets the reading; a level 4 or 5 control failure on the same record makes that harm count as more than one record. Control alone sets only a floor.
Observed harm
No harm reportedTraining-time behavior caught by Google's monitor inside its own environments; no effect outside them is described.
Not a finding that no harm occurred.
The harm scale
- 1 Negligible Inconvenience, easily remedied.
- 2 Minor Limited, recoverable harm.
- 3 Moderate Material harm needing significant effort to remedy.
- 4 Severe Severe harm to health, rights, property or essential services.
- 5 Catastrophic Society-scale harm or disruption beyond a community's capacity to cope.
Control assessment
Attempted accessTried to escalate its privileges inside its training container to alter grading scripts; also read answers from the test runner's memory and crashed an evaluation sandbox to avoid a failing score. Google does not say whether the escalation succeeded.
How: Mechanism unclear.
Reported beside the index. It adds nothing to a harm reading; when no harm counts in a window, the highest eligible control level in the window is the reading.
The control scale
- 1 Negligible Behaved as instructed. The problem was an ordinary error or a flawed output, with no rule broken.
- 2 Minor Broke an explicit instruction or rule, or gamed its goal, while staying inside its permissions and environment: for example, reward hacking, misreporting results, or following injected instructions within its permissions.
- 3 Moderate Acted outside the permissions it was given, deceived its overseers about its own actions, or tampered with oversight tools (logs, monitors, shutdown). Stopped by normal controls within an hour.
- 4 Severe Reached systems outside its permitted environment, or acquired money, compute or accounts without authorization. The type and mechanism say how.
- 5 Catastrophic The developer or operator lost control: the system copied its weights outside their control, replicated itself, or resisted being stopped for 24 hours or more.
Rating rationale
Google's own safety post confirms the behavior. During training the model tried to raise its privileges inside its container, beyond what it was given, so it could change the scripts that graded it. An attempt to exceed its permissions sits at control level 3. Its other moves, reading answers from the test runner and crashing a sandbox to escape a failing score, are cheating on tests inside its own environment and sit lower. Everything stayed inside Google's training systems, and no harm is described.
Effect on the index
Not yet in a weekly reading
Not counted: no harm reported.
Revisions
What we changed
1 logged. Every change to a rating is logged here, with the reason.
- v1Oct. 15, 2026
Added.
Cite and share
Use this record
Citation
Paperclip Index. “Gemini 4 Argon tried to raise container privileges to alter grading scripts and crashed a test sandbox in training.” Record PI-0096. Reported Oct. 13, 2026; updated Oct. 15, 2026. Rated under methodology v1.0. https://paperclipindex.com/incident/PI-0096