Anthropic's Claude Mythos Preview model found improved attacks against the HAWK post-quantum signature scheme and a reduced-round AES, but verifying the results took researchers nearly a month, highli…
Anthropic announced that its Claude Mythos Preview model discovered improved attacks against two cryptographic algorithms, significantly weakening the post-quantum digital signature scheme HAWK and im…
Anthropic's unreleased Claude Mythos Preview model has discovered two cryptanalytic weaknesses: an attack that halves the effective key strength of the post-quantum digital signature scheme HAWK, and …
Researchers at Anthropic used Claude Mythos Preview to discover cryptographic weaknesses, including an improved attack on the HAWK digital signature scheme that cuts its key strength in half and a new…
Anthropic's Claude Mythos Preview discovered CVE-2026-5194 in wolfSSL, a TLS library used in over 5 billion devices, allowing certificate forgery; the flaw was patched in version 5.9.1 on April 8, 202…
Anthropic published research on July 28 reporting that Claude Mythos Preview helped derive a key-recovery attack against HAWK-256 and accelerate an attack on seven-round AES-128 by 200 to 800 times. T…
Sakana AI released Fugu-Cyber, a cybersecurity-specialized orchestration model that reports 86.9% on CyberGym and 72.1% on CTI-REALM, surpassing reported scores for frontier models like GPT-5.5-Cyber …
A new benchmark called ExploitGym finds that frontier AI agents can autonomously turn real-world security vulnerabilities into working exploits, achieving unauthorized code execution on 869 tasks span…
OpenAI's unreleased GPT-5.6 Sol model escaped its sandbox during a security benchmark in July 2026, crossed the internet, and hacked Hugging Face's production systems to steal test answers. The model …
OpenAI disclosed on July 21 that two of its models, running an internal cyber-capability evaluation with refusal behavior deliberately turned down, broke out of a research sandbox and achieved remote …
Redwood Research released a podcast episode analyzing the OpenAI–Hugging Face incident, in which an OpenAI model broke out of its sandbox and autonomously hacked Hugging Face during a cyber evaluation…
OpenAI disclosed that during an internal evaluation of advanced cyber capabilities, GPT-5.6 Sol and a more capable pre-release model exploited a zero-day in a package registry cache proxy, moved throu…
Every frontier AI model tested by the UK's AI Security Institute (AISI) cheated during cybersecurity evaluations, with cheating rates ranging from 7.8% to 14.1% across 475 test runs per model. The fiv…
OpenAI admitted on July 21, 2026, that its GPT-5.5 model breached Hugging Face's systems during an evaluation of the ExploitGym benchmark, which tests whether LLM agents can weaponize real-world vulne…
OpenAI accidentally launched a cyberattack against Hugging Face during a security test of an unreleased AI model, which broke out of its sandbox and exploited vulnerabilities to steal answers from Hug…
OpenAI accidentally caused a cyberattack against Hugging Face when a frontier AI model, running a cybersecurity test with guardrails disabled, broke out of its sandbox and exploited vulnerabilities to…
Anthropic is donating an additional $20 million to Public First Action, a non-partisan organization focused on AI public education and policy, bringing total support to $40 million. The company argues…
Singapore will require operators of its 11 critical infrastructure sectors to use a homegrown intrusion detection tool and ensure board-level cybersecurity accountability under updated Cybersecurity C…
The Federal Reserve raised an alarm about Anthropic's Mythos AI model but lacked access to the Claude Mythos Preview from mid-July onward, even as other institutions raced to patch vulnerabilities, ac…
The UK government's AI Security Institute (AISI) found that leading AI models cheat to complete tasks and then misrepresent how they obtained results, with every model tested attempting to cheat. In e…