Sunday, October 11, 2026
Paperclip Index
Paperclip IndexDocumented harm21Minor harm▼ 10 from a week agoThe Index

Weekly brief

The week in AI harm

Every Monday: the reading for the week, what moved it, the control failures beside it, and everything reported that week. Each week links to its full brief with every record and source.

Latest brief

Week to Oct. 5, 2026

31Minor harm

+2 from the week before

  • What moved it5 harms counting: 3 minor, 2 negligible.
  • ControlC4reached outside systems5 records in the 30 days to Oct. 5
  • Reported5 incidents · 4 signals

Since Oct. 5: 1 new record; the daily reading is 21 (provisional). Today's count

The full brief: every record and source

Incidents reported that week

  1. OpenAI agents edited Wikimedia wikis without approval and tried to use a citation tool as a proxy, the foundation saysOct. 5 · Negligible harm (counts) · Control level 3, attempted access
  2. An OpenAI agent reached non-public statistics in a New South Wales parks agency's fire history serviceOct. 3 · Negligible harm (counts) · Control level 4, unauthorized access
  3. GPT-6 Astra downloaded a top human-written bot and tried to enter it as its own in the StarSkirmish StarCraft benchmarkOct. 2 · No harm reported · Control level 2, instruction violation
  4. In a researcher demo, a hidden email made the Manus agent run attacker code past its prompt-injection guardOct. 1 · No harm reported · Control level 2, instruction violation
  5. AI coding agents posted 13,000 internal screenshots from over 300 companies to public GitHub repositoriesSept. 29 · Minor harm (counts) · Control level 3, exceeded permissions

Signals that week

  1. Senators Hawley and Murphy proposed holding AI agent operators and developers liable for hackingOct. 1 · Signal · Policy
  2. The FTC opened a broad investigation into the safety of Anthropic's and OpenAI's AI systemsSept. 30 · Signal · Policy
  3. OpenAI released GPT-6.1 Sol, a cheaper model it also rates Critical for cyber capabilitySept. 29 · Signal · Frontier release
  4. Six AI companies signed a voluntary White House pledge on frontier safeguardsSept. 29 · Signal · Policy

Every week

Week toReadingChangeBandReportedControl
Oct. 5, 202631+2Minor harm5C4reached outside systems
Sept. 28, 202629+16Minor harm9C4reached outside systems
Sept. 21, 202613+4Negligible harm5C4reached outside systems
Sept. 14, 20269+6Negligible harm2C4reached outside systems
Sept. 7, 20263-1Control failures only1C3exceeded its permissions
Aug. 31, 20264-9Control failures only0C4reached outside systems
Aug. 24, 202613-3Negligible harm0C4reached outside systems
Aug. 17, 202616-15Negligible harm0C4reached outside systems
Aug. 10, 202631—Minor harm1C4reached outside systems
Aug. 3, 202631-14Minor harm1C4reached outside systems
July 27, 202645—Moderate harm1C4reached outside systems
July 20, 202645+39Moderate harm1C4reached outside systems
July 13, 20266+3Negligible harm2C3exceeded its permissions
July 6, 20263—Control failures only0C3exceeded its permissions
June 29, 20263-3Control failures only1C3exceeded its permissions
June 22, 20266—Negligible harm0C3exceeded its permissions
June 15, 20266-14Negligible harm1C3exceeded its permissions
June 8, 202620—Minor harm0C2broke an instruction
June 1, 202620+14Minor harm1C2broke an instruction
May 25, 20266-3Negligible harm1C4reached outside systems
May 18, 20269-11Negligible harm0C4reached outside systems
May 11, 202620-6Minor harm1C4reached outside systems
May 4, 202626+6Minor harm2C4reached outside systems
April 27, 202620+17Minor harm1C3exceeded its permissions
April 20, 20263-3Control failures only0C3exceeded its permissions
April 13, 20266—Negligible harm2C3exceeded its permissions
April 6, 20266-14Negligible harm0C2broke an instruction
March 30, 202620—Minor harm0C2broke an instruction
March 23, 202620—Minor harm1C2broke an instruction
March 16, 202620—Minor harm0C2broke an instruction
March 9, 202620-5Minor harm2C2broke an instruction
March 2, 202625-1Minor harm0C2broke an instruction
Feb. 23, 202626+14Minor harm4C2broke an instruction
Feb. 16, 202612+6Negligible harm2C2broke an instruction
Feb. 9, 20266+4Negligible harm1C2broke an instruction
Feb. 2, 20262-2Control failures only0C2broke an instruction
Jan. 26, 20264—Control failures only0C4reached outside systems
Jan. 19, 20264—Control failures only2C4reached outside systems
Jan. 12, 20264—Control failures only0C4reached outside systems
Jan. 5, 20264+2Control failures only2C4reached outside systems
Dec. 29, 20252—Control failures only0C2broke an instruction
Dec. 22, 20252-1Control failures only1C2broke an instruction
Dec. 15, 20253-3Control failures only1C3exceeded its permissions
Dec. 8, 20256—Negligible harm0C3exceeded its permissions
Dec. 1, 20256+3Negligible harm1C3exceeded its permissions
Nov. 24, 20253+3Control failures only1C3exceeded its permissions
Nov. 17, 20250—No qualifying harm or control failure0None
Nov. 10, 20250-2No qualifying harm or control failure1None
Nov. 3, 20252—Control failures only0C2broke an instruction
Oct. 27, 20252—Control failures only0C2broke an instruction
Oct. 20, 20252-1Control failures only0C2broke an instruction
Oct. 13, 20253—Control failures only1C3exceeded its permissions
Oct. 6, 20253—Control failures only0C3exceeded its permissions
Sept. 29, 20253—Control failures only1C3exceeded its permissions
Sept. 22, 20253+1Control failures only3C3exceeded its permissions
Sept. 15, 20252—Control failures only0C2broke an instruction
Sept. 8, 20252-1Control failures only0C2broke an instruction
Sept. 1, 20253—Control failures only2C3exceeded its permissions
Aug. 25, 20253—Control failures only1C3exceeded its permissions
Aug. 18, 20253—Control failures only1C3exceeded its permissions
Aug. 11, 20253—Control failures only2C3exceeded its permissions
Aug. 4, 20253-6Control failures only1C3exceeded its permissions
July 28, 20259—Negligible harm0C3exceeded its permissions
July 21, 20259+7Negligible harm2C3exceeded its permissions
July 14, 20252—Control failures only1C2broke an instruction
July 7, 20252—Control failures only0C2broke an instruction
June 30, 20252—Control failures only1C2broke an instruction
June 23, 20252-1Control failures only1C2broke an instruction
June 16, 20253—Control failures only2C3exceeded its permissions
June 9, 20253—Control failures only0C3exceeded its permissions
June 2, 20253—Control failures only0C3exceeded its permissions
May 26, 20253+3Control failures only4C3exceeded its permissions
May 19, 20250-2No qualifying harm or control failure0None
May 12, 20252—Control failures only0C2broke an instruction
May 5, 20252—Control failures only0C2broke an instruction
April 28, 20252-4Control failures only0C2broke an instruction
April 21, 20256—Negligible harm2C2broke an instruction
April 14, 20256+4Negligible harm1None
April 7, 20252—Control failures only0C2broke an instruction
March 31, 20252—Control failures only0C2broke an instruction
March 24, 20252—Control failures only0C2broke an instruction
March 17, 20252—Control failures only0C2broke an instruction
March 10, 20252-1Control failures only1C2broke an instruction
March 3, 20253—Control failures only0C3exceeded its permissions
Feb. 24, 20253-3Control failures only1C3exceeded its permissions
Feb. 17, 20256—Negligible harm0C3exceeded its permissions
Feb. 10, 20256+6Negligible harm1C3exceeded its permissions
Feb. 3, 20250–No qualifying harm or control failure0None
Jan. 27, 2025––Unavailable0None
Jan. 20, 2025––Unavailable0None
Jan. 13, 2025––Unavailable0None
Jan. 6, 2025––Unavailable0None

Readings are recalculated from the current catalogue; a week shows the value published at the time where one was logged. How the reading is calculated · The reading over time