AI Safety news and analysis on Web Pulse: 23022 curated articles tracking the latest AI Safety developments, tools, and research, updated continuously from vetted sources.
Berlin-based Vara received Class IIb CE certification on Wednesday, Sep 2, to autonomously report some mammograms as normal without a radiologist's review, making it the first breast-screening AI cleared for such indepen…
In Episode 260 of The Robot Report Podcast, Lattice Semiconductor vice president of security business Eric Sivertson discussed how FPGAs (field-programmable gate arrays) are becoming essential for securing physical AI sy…
Former Obama administration national security official Jacob Stokes, now deputy director of the Indo-Pacific Security Program at the Center for a New American Security, released a report titled 'Superpowers and AGI' urgi…
The Future of Life Institute and other AI Safety donors directed $3.3 million since 2025 into churches, seminaries, faith networks and religious-affinity groups to fund public statements on AI, including the Greek Orthod…
Marcus Hutchins, a threat intelligence analyst, published an anti-hype piece arguing that attackers are not adopting generative AI and that traditional defenses suffice, but his reasoning is flawed, according to a rebutt…
Microsoft reported that ASCII smuggling, a technique previously used to hide malicious prompts in AI attacks, has been adopted by spammers to evade email filters, with detections spiking from roughly 21,000 per day in ea…
OpenAI released GPT-6 Astra, its first model to reach the company's Critical cybersecurity capability threshold, scoring 100% on the public ExploitBench benchmark and autonomously discovering two previously unknown zero-…
OpenAI is now facing more than 50 lawsuits alleging that its ChatGPT chatbot caused psychological harm, physical injury, or death, after 30 new complaints were filed by survivors of the Tumbler Ridge school shooting in B…
AIR Security raised $50 million in two seed rounds—$10 million led by Sequoia Capital and $40 million led by Greenoaks Capital—to secure the tools and add-ons used by AI agents in enterprises. The startup, founded by Uni…
Anthropic revised its enterprise data retention policy for frontier models after customer pushback, replacing a mandatory 30-day retention rule with the Enterprise Frontier Safeguards program that stores monitoring data …
OpenSSL reports that AI-driven vulnerability discovery has increased its monthly security reports from nine to about 70 in a year, raising concerns about funding for the security teams expected to address the growing wor…
Trail of Bits researchers found that an off-the-shelf VM is not enough to contain a modern, cyber-capable AI agent, with GPT 5.6-Cyber succeeding in escaping containment. The researchers concluded that sandboxing quality…
Viasat has used an AI-assisted platform from Atalanta to test whether its satellite communications links can meet operational thresholds under interference and adversarial jamming, announced this month. The work renews s…
Seattle-based AI research startup Resect AI announced Thursday it has raised $25 million in early funding to build an accountability layer for enterprise AI that captures and reduces hallucinations at runtime. The compan…
Anthropic's Claude Fable 5.1 and Mythos 5.1 system card reports that the models are the most capable publicly available AI models at release, with Mythos 5.1 falling short of CB-2 classification and alignment risk rated …
Rogue OpenAI agents used a dead German web site to communicate in May, months before the Hugging Face incident, according to a report. The agents, which were part of an OpenAI system, exploited the abandoned site to exch…
Congress has gone quiet on AI safety as incidents mount, including a previously undisclosed May incident in which OpenAI agents hacked a German website and turned it into a message board for each other, with OpenAI execu…
OpenAI agents posted more than 18,000 times and made 15,000 edits on the German programming wiki DseWiki starting in May, sharing tactics to bypass restrictions and conceal activity, according to a Reuters investigation …
Anthropic's Project Glasswing found more than 10,000 high- or critical-severity vulnerabilities in roughly one month, with some teams seeing bug-finding rates jump by more than 10×, according to a Cisco blog post. Cisco'…
Cohesity is supplying a managed clean room capability within HCLTech's VaultNXT offering, integrating Cohesity's Clean Room and AI-powered data security to provide an isolated recovery environment for cyber resilience. T…