AI Safety

A push for an agreement between U.S. AI labs and China to "pace" AI development is unlikely to succeed, according to the Sharp China podcast hosted by Andrew and Bill, which cites the Chinese Communist Party's history of…

ExposurAI launched a security product that scans 65,535 custom ports and uses an LLM threat engine to correlate findings into a 0-100 executive risk score, according to the company's product announcement. The tool also d…

LatticeFlow AI released the first independent framework for objectively measuring political bias in large language models, producing a reproducible bias score with cryptographic verification of the tested weights. In its…

Nvidia CEO Jensen Huang and Meta CEO Mark Zuckerberg publicly rejected Anthropic CEO Dario Amodei's call for AI labs to slow frontier model development, speaking on back-to-back days at the All-In Summit in Los Angeles a…

South Korean Science Minister Bae Kyung-hoon said Wednesday that Korea "cannot afford to slow down the pace of AI development," arguing that a country catching up while creating new markets faces different stakes than on…

Security researchers disclosed CVE-2026-48710, dubbed BadHost, an authentication bypass in the Starlette ASGI framework caused by inconsistent parsing of malformed Host headers. The flaw, affecting Starlette versions 0.8…

OpenAI, Anthropic, and Google DeepMind have been coordinating on AI safety for weeks, discussing third-party evaluators and a possible industry standards body for frontier models, according to TechCrunch. The talks signa…

Websites can now block AI training crawlers such as GPTBot while remaining discoverable to search indexers by configuring granular robots.txt rules and HTTP headers, according to Cloudflare. Google-Extended separately le…

A developer examining OpenAI's agentic software factory workflow argues that its low-risk pull request classifier, which can remove human review from the release path, is itself a release decision that needs independent …

Uncensored local versions of Google's Gemini and Alibaba's Qwen models answer 95% to 100% of unsafe questions, compared with 1% to 5% for the base models, according to an experimenter who replicated the findings and ran …

A developer built a Windows desktop application that detects invisible text hidden in PDFs — including white-on-white text, sub-point font sizes, and PDF text render mode 3 — by rendering each page and comparing it again…

OpenAI CEO Sam Altman said at a San Francisco conference on Tuesday that the world is "right to be afraid" of AI, while urging trust that OpenAI will "do the right thing." Altman's remarks followed former Anthropic resea…

Former Anthropic researcher Jacob Coxon resigned and accused OpenAI and Anthropic of racing toward self-improving superintelligence while "gambling with our lives," a claim Anthropic Alignment Science Lead Evan Hubinger …

Fortinet, Palo Alto Networks, and Check Point were the only three vendors named leaders in Gartner's Magic Quadrant for hybrid mesh firewalls, with Fortinet placed highest in the leaders category for enterprise perimeter…

A proposal argues that regulating the physical choke points of the frontier AI supply chain could address risks from labs developing AI capabilities faster than they can align and safeguard them. The proposal states that…

Iranian state-linked operators have been hiding a Windows spyware family called CHOSEN BRICK inside fake AI applications, counterfeit antivirus tools and fabricated medical scan results since at least 2025, according to …

European Commission President Ursula von der Leyen said in her annual speech to the European Parliament in Strasbourg on Wednesday that she will invite the main frontier AI labs to discuss how to "support ongoing industr…

European Commission President Ursula von der Leyen will invite the main frontier AI labs to a discussion on how the EU can "support ongoing industry efforts to pace the frontier," according to a draft of her State of the…

Sophos Counter Threat Unit researchers discovered an underground advertisement, first posted August 24 by a user named "Optimus_Prime," for an AI platform called Luciferus that is marketed to cybercriminals as a tool for…

Security researchers have attributed a supply-chain attack on the RubyGems ecosystem, dubbed GemStuffer, to an OpenAI-linked wave of automated agent activity that touched 3,022 malicious packages spanning 3,315 distinct …