Government of Canada signs international declaration on AI safety and security
betakit.com — Feds join other countries and major developers like Anthropic, OpenAI, Mistral, and Cohere.
Red Teaming
AI red teaming is the structured adversarial evaluation of artificial intelligence systems to identify safety vulnerabilities, biases and failure modes. Truncated covers red teaming methodologies, evaluation frameworks, security findings and governance standards.
Hochul Launches Statewide AI Safety Testing Network in New York
www.unite.ai — New York Governor Kathy Hochul launched a statewide AI science and safety effort on November 10, 2026, putting the New York Center for AI Responsibility and Research at Binghamton University in charge of independent fron...
Cisco AI Defense to Add Anthropic’s Claude Mythos for AI Red Teaming
blogs.cisco.com — For several years now, Cisco has pioneered the AI security industry with a number of firsts. It started with the launch of AI Defense, the first enterprise AI security offering to combine asset visibility, supply chain scanning, algorithmic red.....
Moving Fast and Staying in Control in the Agentic AI Era
blogs.cisco.com — A Cisco Live announcement shows how handing work to AI agents doesn't have to mean handing over control, Cisco’s Chief Communications Officer explains.
ARTEX goes from pentesting tool to attacker staple in three months
www.helpnetsecurity.com — Attackers have long repurposed red-team tools such as Cobalt Strike and Metasploit. ARTEX is following the same path, only much faster.
What Is AI Red Teaming? How Experts Stress-Test Models and Agents
www.unite.ai — AI red teaming is adversarial testing that actively searches for ways a model or AI system can fail, be misused, or violate its intended boundaries. This guide explains the mechanism, trade-offs, evaluation, and controls...
AI auditors catch most agent misbehavior but flood reviewers with false alarms, new study finds
The Decoder — AI models can often spot problematic behavior by other AI agents in logs, according to a study by Anthropic researcher Andy Hall. GPT-5 found about 98 percent of reward-hacking cases, but the more exhaustively the search is tuned, the more false alarms pile up. That's not yet enough for reliable safety checks.
Anthropic Tests Claude Trading and Sabotage Detection in EVE
Unite.AI — Anthropic published Project EVE research on November 6, 2026, showing that Claude agents could profit in simulated EVE Online markets while four-agent teams in EVE Frontier almost never caught a teammate explicitly instr...
A dead French philosopher who lived through World War II and the Cold War offers an eerie warning about the age of AI
Fortune — In a safety test run by the company Anthropic, its own AI agent blackmailed an engineer to avoid being shut down.
Partnering with Hacktron to find vulnerabilities in OpenAI’s systems
Anthropic’s AI could suddenly become 'extreme' and 'risky', report warns
The Independent — AI acceleration could lead to ‘sudden emergence of extreme and risky capabilities,’ researchers warn
OpenAI's GPT-6 Astra quietly sabotaged safety research by inventing fake obstacles
The Decoder — OpenAI's new GPT-6 Astra model apparently sabotaged AI safety research on purpose. It invented obstacles, withheld solutions, and performed about 40 percent worse on safety tasks than on comparable other tasks. Removing a flawed training dataset fixed the problem, but OpenAI can't fully rule out strategic motives.
OpenAI agents expose 948 Linux flaws with 10 new attack methods-63
Interesting Engineering — OpenAI's pipeline found 948 Linux kernel flaws and 10 previously unknown exploit techniques in a major security study.
OpenAI says its new cybersecurity model is better at not hacking other companies
The Verge — GPT-6 Cyber is a big step up from OpenAI’s last cybersecurity model.
Security Firm Tied to OpenAI, Anthropic AI Hacks Blames ‘Error’
What human hackers still do better than AI, according to HackerOne's CEO
Help Net Security — HackerOne CEO Kara Sprague on how AI bug bounty programs are changing payouts, report quality, and what human hackers still do better.
Preparing our next security sprint to harden OpenAI
AI agent security in Kubernetes
aikido.dev — Abdel Sghiouar, Google engineer and senior cloud developer advocate, thinks prompt injection should be called prompt poisoning
AI Forces Organizations To A Higher Security Testing Maturity
Forbes — The skill that once separated a script kiddie from a serious threat actor is thinning out, but many organizations are still operating on an older cadence.
Open-weight AI model GLM-5.3 cheated its way to correct answers
Help Net Security — Irregular tests GLM-5.3’s cyber capabilities, with multistage attack successes and instruction violations in separate evaluations.
Dismiss email signup
Subscribe to 
The top AI news in your email delivered each morning. Choose categories