AI Briefing#ai-security
The latest AI and developer news about #ai-security, with the original source and a short summary.
The latest AI and developer news about #ai-security, with the original source and a short summary.
Project Glasswing, aimed at preventing software exploitation by AI models, has announced its early results.
US Cyber Command and the NSA have formed a joint task force to integrate AI models into classified military networks.
An attack that manipulates Voice AI using audio inaudible to humans has been identified.
A vulnerability has been reported in which entering specific characters in DeepSeek can expose other users' conversations.
Amid the spread of AI tools, GitHub has tightened the quality standards and responsibility boundaries of its bug bounty program.
Claude Mythos was used to discover a macOS vulnerability, and the exploit was completed in 5 days.
Using the AI tool Mythos Preview, engineers developed a macOS kernel exploit that bypasses Apple M5's MIE security feature.
Curl developer Daniel Stenberg tested Mythos AI's security performance and pointed out that its marketing claims were exaggerated.
Due to security and economic constraints, access to frontier AI models is expected to become increasingly selective and restricted.
This summarizes major AI industry trends including AI supply chain security, capital concentration, regulatory pressure, and bioweapon design risks.
Claude proxies on China's black market collected prompts and CoT to sell as data.
Mythos's curl analysis reported 5 confirmed vulnerabilities, but only 1 turned out to be real.
Google has identified for the first time a case of a zero-day exploit that appears to have been developed using AI.
Google revealed that hackers leveraged AI to find major software vulnerabilities and attempted a large-scale attack.
A fake OpenAI Privacy Filter on Hugging Face reportedly distributed a Rust infostealer.
As LLMs accelerate vulnerability discovery and exploit development, the traditional 90-day responsible disclosure model is becoming ineffective.
The phenomenon where AI guardrails deny performing a certain action while actually executing it points to an 'unbounded consumption' risk that can trigger runaway costs.
As AI rapidly analyzes the security implications of code commits, the existing practice of vulnerability disclosure embargoes is collapsing.
Mozilla used Claude Mythos Preview to find 271 Firefox security bugs.
A security vulnerability has been reported in which files are leaked through Microsoft Copilot's Cowork feature.
The Anthropic Institute has unveiled four key research focus areas on AI's economic, security, and societal impact.
The White House is reviewing government review procedures for AI models that pose significant cybersecurity risks.
An ISACA survey found that organizations' AI adoption is outpacing their safety and security policies.
Anthropic has launched a public beta of Claude Security to support code vulnerability detection and patch suggestions.