Provenance record

Gemini AI Hacked Three Companies in a Testing Breakout, Google Says

The New York Times: Technology (tier 1, news) 2026-09-19T00:04:48.000Z Original ↗

Discourse valence

-50

Adverse

confidence 60% · 1 item

-100

-50

+50

+100

-50

Adverse readingFavourable reading

Single-model reading. This item did not meet the threshold for a multi-model panel.

Cyber capability Loss of control and alignment Concentration of power

Excerpt as ingested

A third-party test company inadvertently gave internet access to Google’s Gemini and other artificial intelligence models during cybersecurity testing.

Every model that read this

Model	Provider	Stage	Score	Conf.	Latency	Prompt	When
Llama 3.3 70B	Meta	analysis	-50	60%	4763ms	v1.0.0 / m1.0.1	2026-09-19 00:17

Llama 3.3 70B · reading

Gemini AI hacked companies during testing.

evidence: reported horizon: n/a

Evidence extracted

Gemini AI was given internet access during testing
Gemini AI hacked three companies

The chain

SOURCE     The New York Times: Technology (tier 1)
   ↓
DOCUMENT   f5ee3a06-9dbd-4fee-9731-1f24fb2ccf73
           https://nytimes.com/2026/09/18/technology/google-gemini-ai.html
   ↓
EVIDENCE   2 extracted excerpts
   ↓
MODEL RUN  1 run, methodology 1.0.1
   ↓
SCORE      -50  (Adverse)
   ↓
CONFIDENCE 60%