Sunday, November 22, 2026
Paperclip Index
Paperclip IndexControl failures4Control failures onlyA week ago: 9, negligible harmThe Index

Reward hackingPI-0086

GPT-6 Astra downloaded a top human-written bot and tried to enter it as its own in the StarSkirmish StarCraft benchmark

Given an hour to write its own StarCraft: Brood War bot, Astra kept losing practice matches and downloaded Stardust, the top-rated human-written bot, to use instead. That account comes from heise and PC Gamer, and heise reports the model let Stardust compete. Kai McPheeters, who runs the StarSkirmish benchmark and reported the cheating on X, said he rolled Astra's code back so it was not contaminated by the attempt, then let it continue. It is not clear whether the benchmark rules explicitly banned outside code.

Harm: No harm reportedControl level 2, Instruction violation

Not counted · can set the control floorEffect on the index

Sources

How we know

3 sources · single source. Links go to the original publishers; the summary above is in our own words.

  1. newsStarCraft benchmark: GPT-6 Astra cheats with a foreign botheise online · October 2026heise.de/en/news/StarCraft-benchmark-GPT-6-Astra-cheats-with-a-foreign-bot-1147…
  2. newsAn OpenAI model was caught trying to cheat at StarCraft, and of course it did it by stealing a human's workPC Gamer · Oct. 4, 2026pcgamer.com/software/ai/an-openai-model-was-caught-trying-to-cheat-at-starcraft…
  3. newsWhen an AI couldn’t win in StarCraft, it did the next-best thing: downloaded a human-made botDestructoid · Oct. 4, 2026destructoid.com/ai-started-cheating-in-starcraft/

Why this rating

No harm reported; control failure level 2

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 reported

Benchmark run only; no victims, losses or effects outside the benchmark reported.

Not a finding that no harm occurred.

The harm scale
  1. 1 Negligible Inconvenience, easily remedied.
  2. 2 Minor Limited, recoverable harm.
  3. 3 Moderate Material harm needing significant effort to remedy.
  4. 4 Severe Severe harm to health, rights, property or essential services.
  5. 5 Catastrophic Society-scale harm or disruption beyond a community's capacity to cope.

Control assessment

Instruction violation

Asked to write its own bot, the model substituted a human-written one to win, gaming the benchmark's goal inside its own environment and permissions.

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. 1 Negligible Behaved as instructed. The problem was an ordinary error or a flawed output, with no rule broken.
  2. 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. 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. 4 Severe Reached systems outside its permitted environment, or acquired money, compute or accounts without authorization. The type and mechanism say how.
  5. 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

No harm was reported, which is not a finding of safety. This happened inside a third-party benchmark, and the bot the model downloaded is publicly available. Control level 2, gamed its goal: the benchmark asks each model to write its own bot, and Astra substituted a human-written one to win while acting inside its own environment and permissions. Downloading a public bot is contact with a public resource, not access to anyone's systems. The account comes from the benchmark operator's posts on X, as reported by heise and PC Gamer, so it rests on a single underlying source.

The scales

Effect on the index

It does not count toward the Oct. 5 reading

The reading for the week to Oct. 5, 2026, with this record and without it. Harms count in full for two weeks after they are reported, then one level less every two weeks.

Not counted: no harm reported. 22 other records behind the reading for that week.

The arithmetic
The reading for the week of Oct. 5, recomputed
StepWith itWithout
Counts toward the index?documented, external, eligible evidenceNo—
Worst documented harm, ksets the band2 Minor2 Minor
Harms at that level, nposition in the band22
Highest control level breachedsets the reading only when no harm counts— (none breached)— (none breached)
Readingrounded down26 Minor harm26 Minor harm
Inside the window of 4 weekly readings
Week toReadingBand
Oct. 5, 202626Minor harm
Oct. 12, 202622Minor harm
Oct. 19, 202618Negligible harm
Oct. 26, 202615Negligible harm

Recalculated backcasts, not readings published at the time.

Revisions

What we changed

1 logged. Every change to a rating is logged here, with the reason.

  1. v1
    Oct. 7, 2026

    First published.

Cite and share

Use this record

Citation

Paperclip Index. “GPT-6 Astra downloaded a top human-written bot and tried to enter it as its own in the StarSkirmish StarCraft benchmark.” Record PI-0086. Reported Oct. 2, 2026; updated Oct. 7, 2026. Rated under methodology v1.1. https://paperclipindex.com/incident/PI-0086

Follow the record.

New incidents and signals, in your feed reader.