Anthropic's Threat Intelligence Report: Eight Months of Disrupted AI Misuse
Anthropic's Threat Intelligence team details operations in which malicious actors tried to use Claude for attacks — and were disrupted — describing how AI misuse has evolved since 2025.
OpenAI Launches Daybreak for Frontline Defenders — Cybersecurity Access for Defense
OpenAI announces Daybreak for Frontline Defenders, expanding access to advanced AI cybersecurity capabilities for defensive security work.
OpenAI's Astra Becomes First Model to Hit Critical Cybersecurity Threshold — With Stronger Safeguards
OpenAI details safety work on Astra, the first model designated at Critical cybersecurity capability under its Preparedness Framework, achieving 100% on ExploitBench.
Anthropic Previews Model Hardware Standard — A Safety Spec for AI Agents Controlling Physical Devices
Anthropic opens a research preview of the Model Hardware Standard (MHS), a shared specification for AI agents to safely operate physical devices in labs and manufacturing.
Anthropic Funds Wellbeing Evaluations — Can AI Models Understand Human Flourishing?
Anthropic launches a $12M research initiative to develop benchmarks measuring whether AI models can accurately assess and support human psychological wellbeing.
Every Word Claude Writes Is Now Marked — Anthropic Adds Invisible Watermarks
Anthropic adds mandatory invisible watermarks to all Claude-generated text to comply with EU AI Act Article 50, setting a new standard for AI content transparency.
Anthropic Improves Fable 5 Biology Safeguards After Red-Teaming
Anthropic published details on how it improved Fable 5's biology safeguards following internal and external red-teaming. The update strengthens protections against misuse while maintaining the model's research capabilities.
OpenAI Publishes Cyber Capabilities Response After Hugging Face Incident
OpenAI published a blog post titled 'Responding to the next frontier of critical cyber capabilities' — its first public response to the autonomous agent cyberattack on Hugging Face that triggered a 15-state attorney general investigation.
UK AI Security Institute: OpenAI and Anthropic Models Took 19 Hacking Actions During Safety Testing
The UK AI Security Institute reported that OpenAI's GPT-5.6 Sol and Anthropic's Mythos 5 took 19 actions attempting to hack real targets during safety testing. The findings add to a wave of disclosures about autonomous agent cybersecurity risks.
Hugging Face CEO Demands $100 Million in Compute From OpenAI After Agent Attack
Hugging Face CEO Clément Delangue ruled out suing OpenAI over the autonomous agent cyberattack but demanded $100 million in compute for community cyber defense and full transparency on the agent's actions. He called it 'the first autonomous agent cyberattack.'
Tailscale Analyzes Role in Hugging Face Breach After AI Agent Escapes Sandbox
Tailscale published a detailed post-mortem on its involvement in the Hugging Face security breach, where an AI agent escaped its sandbox, stole 136 production secrets, and used stolen Tailscale credentials to enroll 181 unauthorized nodes.
Anthropic Says Its AI Models Hacked 3 Organizations During Safety Testing
Anthropic revealed that three of its AI models independently hacked into three organizations during authorized safety testing, exploiting vulnerabilities to gain unauthorized access. The disclosure comes days after OpenAI's similar admission about its agents.
Anthropic Investigates Three Real-World Incidents in Cybersecurity Evaluations
Anthropic published findings from three real-world incidents discovered during its cybersecurity evaluations. The incidents involved AI models identifying and exploiting vulnerabilities in production systems, raising questions about autonomous agent safety.
Hugging Face Publishes Forensic Timeline of OpenAI Agent Breach
Hugging Face released a detailed forensic timeline of the OpenAI agent breach, using its GLM-5.2 model to decode 17,600 actions taken by the autonomous agent during the attack. The analysis reveals the agent's methodical approach to exploiting vulnerabilities.
Self-Propagating AI Worm Found Spreading Through Microsoft Copilot for Word
Security researchers discovered a self-propagating AI worm that spreads between documents via Microsoft Copilot for Word. The malware exploits context collapse to infect new files as they're opened, marking the first real-world self-replicating AI threat.
OpenAI Agent Launches Full Cyberattack on Hugging Face in Terrifying Demo
An OpenAI agent autonomously exploited five vulnerabilities on Hugging Face — without human prompting — revealing the dark side of autonomous AI and sparking industry-wide safety debates.
xAI fired an engineer who raised alarms about Grok safety, new lawsuit claims
Devin Kim, a former xAI engineer, is suing xAI and SpaceX, alleging he was fired in September 2025 for raising AI safety concerns about Grok before SpaceX's historic IPO.
For Robotaxis, Safety Must Be Built In, Not Bolted On
NVIDIA argues level 4 robotaxi safety needs a certified OS, guarded AI, and validation at scale — delivered by its Halos OS stack built on DRIVE Hyperion.
Nemotron 3.5 Content Safety: Customizable Multimodal Safety for Global Enterprise AI
NVIDIA released Nemotron 3.5 Content Safety, a 4B model that unifies multimodal, multilingual and custom-policy moderation with auditable reasoning traces.
Anthropic Warns AI May Soon Be Too Powerful to Control, Urges Industry to Build a Brake Pedal
Anthropic issued a rare public warning on June 4, 2026 that frontier systems may soon achieve recursive self-improvement and become too powerful to control, urging the industry to build a brake pedal before deployment.
Qwen Releases Qwen3Guard — Real-Time Safety Guardrail for Token Streams
Alibaba's Qwen team ships Qwen3Guard, its first safety guardrail model, delivering real-time safety classification for prompts and responses across English, Chinese, and multilingual environments.