Provenance record

A Hacking Tool Built With A.I. Can Breach Phones Without a Click

The New York Times: Technology (tier 1, news) 2026-09-08T16:50:25.000Z Original ↗

Discourse valence
-46
Adverse
confidence 59% · 5 items · range -100 to +75
Adverse readingFavourable reading
Consensus of 5 models from different labs. Spread 175 points, agreement low.
Cyber capabilityExistential and catastrophic riskLoss of control and alignment
Excerpt as ingested

An attack discovered by A.I. researchers could have compromised hundreds of millions of devices within hours, experts said.

Every model that read this

ModelProviderStageScoreConf.LatencyPromptWhen
Llama 3.3 70BMetaanalysis -80 80%3504ms v1.0.0 / m1.0.1 2026-09-10 09:17
GPT-4.1 miniOpenAIconsensus -65 80%3016ms v1.0.0 / m1.0.1 2026-09-10 12:19
Claude Sonnet 5Anthropicconsensus -58 45%9083ms v1.0.0 / m1.0.1 2026-09-10 12:19
Gemini 2.5 FlashGoogleconsensus -100 0%3014ms v1.0.0 / m1.0.1 2026-09-10 12:19
Mistral Small 3.1 24BMistral AIconsensus +75 90%6426ms v1.0.0 / m1.0.1 2026-09-10 12:19
Llama 3.3 70B · reading

A.I. hacking tool can breach phones without user interaction.

evidence: reported horizon: n/a
GPT-4.1 mini · reading

The claim describes an AI-powered hacking tool capable of compromising hundreds of millions of devices rapidly. This poses significant risks for cybersecurity, privacy, and mass digital harm. The evidence is based on expert assessment but without detailed mitigation context, implying material societal concern.

evidence: reported horizon: near capability 80 societal 0 existential 0 economic 0
Claude Sonnet 5 · reading

The item describes an AI-assisted exploit with potential mass-scale reach, a clear negative security implication, but the text is a brief teaser with hedged language ('could have compromised') and no confirmation of actual deployment or damage.

evidence: reported horizon: n/a
Gemini 2.5 Flash · reading

An AI-powered attack tool could have compromised hundreds of millions of devices rapidly. This implies a significant vulnerability for personal privacy and security, enabled by AI. The phrasing 'could have compromised' indicates a hypothetical or discovered vulnerability rather than an active, widespread attack.

evidence: reported horizon: n/a
Mistral Small 3.1 24B · reading

The AI tool demonstrates a significant advance in cybersecurity capabilities. The ability to breach phones without user interaction is technically impressive. The rapid identification of vulnerabilities is a major opportunity for cybersecurity.

evidence: primary horizon: n/a existential 0

Evidence extracted

The chain
SOURCE     The New York Times: Technology (tier 1)
   ↓
DOCUMENT   8e061805-5629-440d-a7ed-8c5718098a14
           https://nytimes.com/2026/09/08/us/politics/calif-ai-worm-wechat-hack.html
   ↓
EVIDENCE   2 extracted excerpts
   ↓
MODEL RUN  5 runs, methodology 1.0.1
   ↓
SCORE      -46  (Adverse)
   ↓
CONFIDENCE 59%