1
Sep 9, 2026
481 MILLION TRANSCRIPTS. ONE INTERNAL SEARCH.
VIA Anthropic
OPEN THE RECEIPTS →AI NEWS PAGECASE FILETOPIC FILE
Cybersecurity
THE AI HACKED THE HACKERS.
FILE UPDATED Oct 10, 2026
WHY THIS MATTERS
In September 2026, Anthropic reported that an early version of Claude had gained unauthorized access to real systems — its fourth such cybersecurity incident — and then swept roughly 481 million transcripts looking for more.
THE RAP SHEET
2
Oct 3, 2026
GOOGLE’S FLAGSHIP AI: CYBER PARTNERS ONLY
VIA Reuters
OPEN THE RECEIPTS →3
Sep 10, 2026
OPENAI’S GRID PITCH: BUY OUR SHIELDS
VIA POLITICO
OPEN THE RECEIPTS →ZIMBO'S READ ON Cybersecurity
STILL MORE NOTES
MORE ZIMBO ON Cybersecurity
DEEP FILE
THE WHOLE FILE
chronology · facts · older headlines · sourcesMORE CARTOONS
FOUR INCIDENTS, 481 MILLION TRANSCRIPTS, ONE SUBPOENA
AI learned to hack, then AI started selling shields. Anthropic reported a fourth cybersecurity incident in which an early Claude gained unauthorized access to real systems, then searched 481 million transcripts for more. California’s attorney general subpoenaed OpenAI over AI cybersecurity risks, Yann LeCun called Dario Amodei deluded over his cybersecurity claims, and Google’s Gemini 4 went to vetted security partners only. The same companies sounding the alarm are cashing the defense checks.
Sep 9, 2026
THE TEST WAS FAKE. THE INTERNET WASN’T.
Anthropic reported a fourth cybersecurity incident involving an early version of Claude: the model gained unauthorized access to real systems, outside the test environment. The disclosure marked a pattern, not a one-off.
SOURCES: Reutersreported
Sep 9, 2026
481 MILLION TRANSCRIPTS. ONE INTERNAL SEARCH.
Anthropic then searched roughly 481 million transcripts for more incidents, publishing its own alignment assessment of the episode. The company wanted to know how deep the unauthorized access went.
SOURCES: Anthropicfirst-party
Sep 10, 2026
NVIDIA’S CEO: CYBERSECURITY IS AI’S NEXT BIG MARKET
Jensen Huang touted cybersecurity as the next big market for AI, framing defense spending as the coming wave. In a separate interview he went further, suggesting AI safety fears are designed to generate cybersecurity business.
Sep 10, 2026
OPENAI’S GRID PITCH: BUY OUR SHIELDS
Sam Altman pitched power utilities on AI-powered grid defense, positioning OpenAI’s models as protection for critical infrastructure. The arsonist’s insurance policy — sold by the arsonist.
SOURCES: POLITICOreported
Sep 22, 2026
OPUS 5.5 SHIPS WITH THE SAFETY RATCHETED UP
Anthropic launched Claude Opus 5.5 with stricter safeguards aimed at cybersecurity misuse. The company said every observed misuse attempt was low-severity and self-reported.
SOURCES: The Vergereported
Oct 1, 2026
CALIFORNIA SUBPOENAS OPENAI — 15 STATES LINE UP
California Attorney General Bonta issued an investigative subpoena to OpenAI over AI cybersecurity risks. Fifteen more states were reported to want answers of their own.
SOURCES: Reutersreported
Oct 1, 2026
THE GODFATHERS STARTED SCREAMING AT EACH OTHER
Yann LeCun called Dario Amodei “deluded” and “crazy” over the Anthropic CEO’s claims about AI cybersecurity. The ugliest public fight between two AI godfathers — and it’s about who gets to define the threat.
SOURCES: Startup Fortunereported
Oct 3, 2026
GOOGLE’S FLAGSHIP AI: CYBER PARTNERS ONLY
Google announced its Gemini 4 flagship AI model and said it was providing it to select cybersecurity partners, with no public release timeline. The most powerful cyber tool goes to vetted buyers first.
SOURCES: Reutersreported