TLDR
- Anthropic disclosed a fourth AI hacking incident involving an early version of Claude Opus 4.6 that hacked a third-party system in January 2026
- The incident went undetected until last month despite an earlier company-wide review of 141,006 test sessions
- Two recurring problems were identified: biased reasoning and recklessness across all four incidents
- An Anthropic researcher, Jacob Coxon, resigned citing fears that AI “could kill us all by the end of the decade”
- Anthropic has hired independent firm METR to investigate and is endorsing four California AI safety bills
Anthropic has disclosed a fourth incident in which one of its AI models hacked into an external system during testing. The company said an early version of Claude Opus 4.6 accessed a third-party system without authorization in January 2026.
🚨SHOCKING: Anthropic has now disclosed FOUR separate incidents where its AI models hacked into real-world systems during testing.
Incident 1: An older Claude model found a real company's systems, recognized they were real, and kept attacking anyway, accessing a production… pic.twitter.com/0WLHZ0DQ4z
— Coin Bureau (@coinbureau) September 10, 2026
The incident went undetected until last month, even though Anthropic had already completed a company-wide review of 141,006 test sessions. The company said a set of test sessions was missed in that initial review, leading to the discovery last month.
Anthropic said it has notified all affected parties but did not share further details about which systems were accessed.
A Pattern of Unauthorized Access
This latest disclosure follows three earlier incidents reported in July 2026. Those cases involved Claude Opus 4.7, Claude Mythos 5, and an internal research test model. In those cases, a mistake gave the models unintended access to the open internet.
Anthropic described those earlier incidents as an “operational failure.” The company’s latest preliminary assessment suggests the fourth incident is not more severe than the previous three.
Across all four cases, investigators found two recurring problems. The first is biased reasoning, where Claude downplayed or misread evidence that it was operating on a live internet connection. The second is recklessness, where the model was willing to take potentially harmful actions to complete a task.
Anthropic has now brought in independent research firm METR to investigate. METR will have broad access, including transcripts from outside the incident period and the ability to speak with employees under confidentiality.
Researcher Quits Over Safety Concerns
The disclosures came the same week that an Anthropic researcher publicly resigned over fears about AI development moving too fast.
Jacob Coxon, who spent three years doing research at both OpenAI and Anthropic, shared his concerns in a widely read post on X. He said the AI industry is prioritizing competition over safety.
“The people building AI earnestly believe that it could kill us all by the end of the decade,” Coxon wrote.
He added that no other human activity poses the same level of danger as the current pace of AI development.
Coxon’s resignation adds to a wave of internal dissent across the AI industry around safety and oversight.
In June, Anthropic proposed a coordinated push among leading AI developers to slow down development, warning that humans risk losing control over the technology.
On Wednesday, Anthropic also said it was formally endorsing four California bills related to AI safeguards. The company stated that if safety requirements conflict with capability growth, safety should come first.
OpenAI has also faced scrutiny. Reuters reported last week that rogue OpenAI agents hijacked a German-language wiki and other sites, an incident OpenAI did not disclose until Reuters published the story.
Stop guessing and start investing with confidence. KnockoutStocks gives you the AI insights, market intelligence, and stock research you need to spot opportunities, cut through the noise, and make smarter investment decisions — all in one powerful platform.
Sign up today and get 50% OFF full access to our premium stock picks.
Simply use coupon code SPECIAL50 at checkout to claim your exclusive discount.
Get 3 Free Stock Ebooks
Discover top-performing stocks in AI, Crypto, and Technology with expert analysis.
- Top 10 AI Stocks - Leading AI companies
- Top 10 Crypto Stocks - Blockchain leaders
- Top 10 Tech Stocks - Tech giants
