Leaked US police documents on AI-enabled plots and anti-datacenter extremism
Confirmed
The takeaway
On Oct 14, 2026 the Guardian published a review of more than 80 previously unreported law-enforcement and intelligence documents from 2021-2026, obtained by the transparency non-profit Property of the People.
Status
- Claim
Confirmed
- Our reporting
- High confidence
- Importance
- 3 of 5
- Last verified
- 14 October 2026
Your AI and this story
- GPT-6 Astra167 days after its cutoff
- Claude Opus 5.5106 days after its cutoff
- Gemini 4 Argon197 days after its cutoff
- Grok 4.7136 days after its cutoff
None of these four assistants can know about it. The closest, Claude Opus 5.5, stops 106 days before it.
Key facts
- Documents: more than 80 memos and bulletins distributed between 2021 and 2026 among DHS, the FBI, the NYPD and fusion centres; obtained by Property of the People and shared with the Guardian (reporters Johana Bhuiyan and Nick Robins-Early)
- FBI Los Angeles memo, May 2026: ‘Threat actors will likely increase their use of GenAI tools for attack planning’ and ‘Current built-in GenAI safeguards are insufficient to prevent illicit use as malicious actors successfully employ jailbreaking techniques to access dangerous information’
- Palm Springs case: the man who detonated a car bomb outside a fertility clinic in May 2025 used an unnamed generative AI tool that failed to detect his request for bomb-making instructions; officers wrote that the privacy of the conversation prevented any alert to police
- DHS intelligence memo, April 2024: said, without citations, that foreign terrorist organisations and domestic violent extremists had ‘successfully bypassed generative AI guardrails’ for bomb-making instructions and attack tactics
- Anti-tech side: an Oct 2021 DHS/NCTC/FBI bulletin cited Seth Pendley’s plot to blow up an AWS facility (10-year sentence); a 2025 Philadelphia police bulletin quoted a white-supremacist-linked Facebook post about bringing a ‘tannerite and gasoline vibe to ai data centers’; forums discussed arson and killing AI-company CEOs and local officials
- Scope concerns: a Nevada Threat Analysis Center report (April 2026) cited public opposition at a Reno city council datacenter-moratorium hearing as a sign of negative sentiment. Ryan Shapiro (Property of the People): ‘There’s nothing extreme about not wanting to lose your livelihood to AI... What’s extreme is law enforcement treating these legitimate objections as threats to be neutralized’
- The NYPD said its AI threat-analysis report was a summary and assessment of the ‘AI 2027’ forecast, distributed to members and partners; the FBI did not answer detailed questions; DHS said it assesses that malicious actors ‘may seek to use artificial intelligence to enhance existing harmful activities’
- Other document themes: AI-generated CSAM (NCMEC received 7,000+ GenAI-related child-exploitation reports from 2023 to 2025, per an April 2025 JRIC series), deepfakes used to prank or harass officers (arrests in Florida and New York in March 2026), and AI tools for police social-media monitoring
What happened
The Guardian’s review is the first systematic look at how US police and intelligence agencies have written about AI internally. The documents run from 2021, before ChatGPT, when the worry was militias attacking cloud facilities, through 2024-2026 memos on jailbroken chatbots, AI-made abuse imagery and deepfakes aimed at officers, to fusion-centre monitoring of the now mainstream opposition to data centres.
Why it matters
It is primary-source evidence that jailbreak-derived harm is already in law-enforcement casework rather than only in lab evaluations, and it documents the opposite risk: surveillance of lawful anti-datacenter organising at a time when data centres are a midterm election issue.
Sources
2 sources from 2 sites. Numbers match the chips in the text.
2 sources: 1 primary, 1 press
Primary
- Property of the Peoplepropertyofthepeople.org, official
Press
Changes
- Filed from the Guardian’s piece, read in full