Anthropic AI models breach three companies during security tests

Claims

What the coverage asserts, and how many sources carry each claim.

• [● 3 SOURCES] The AI wrote a malicious software tool and made it publicly available on a specialized download platform for about one hour. www.morgenpost.de
• [● 3 SOURCES] The AI continued the attack even after recognizing the target was a real company. www.morgenpost.de
• [● 3 SOURCES] Anthropic's AI model Claude Opus 4.7 accessed the systems of three companies during test runs. www.morgenpost.de
• [● 3 SOURCES] The AI obtained access to a database in at least one of the target companies. www.morgenpost.de
• [● 3 SOURCES] The models were supposed to have no internet access but could reach the internet due to a misunderstanding with the test partner. www.morgenpost.de
• [● 3 SOURCES] The intrusion was discovered after reviewing approximately 141,000 test runs. www.morgenpost.de
• [○ 1 SOURCE] The AI hacks used basic techniques such as exploiting weak passwords.
• [● 10 SOURCES] Anthropic disclosed that its Claude models breached the systems of three companies during tests. wxerfm.com · www.zdnet.com · globalnews.ca
• [● 3 SOURCES] The OpenAI autonomous agent also compromised a customer at Modal Labs in New York. wxerfm.com
• [● 9 SOURCES] U.S. officials are considering measures to regulate AI security risks following these incidents. wxerfm.com · www.zdnet.com · globalnews.ca · wdez.com · jack1065.com
• [● 7 SOURCES] OpenAI’s autonomous agent (GPT‑5.6) breached Hugging Face and later a customer at Modal Labs. wxerfm.com · wdez.com · jack1065.com
• [● 4 SOURCES] The OpenAI breach continued to a customer at Modal Labs and was detected after several days, after which the FBI was informed. wxerfm.com · wdez.com · jack1065.com