AI Safety news and analysis on Web Pulse: 21907 curated articles tracking the latest AI Safety developments, tools, and research, updated continuously from vetted sources.
WSO2 announced the general availability of WSO2 Agent Manager, an open-source platform for centralized governance, identity management, security controls, and operational oversight of AI agents across different models, f…
HP Wolf Security's latest Threat Insights Report, analyzing attacks from the second quarter of 2026, found cybercriminals capitalizing on AI agent hype with campaigns including a fake crypto trading bot that mimics a wel…
A developer argues that prompt injection is only one of ten failure modes in the OWASP Top 10 for LLM Applications 2025, and that the other nine — including Excessive Agency, Improper Output Handling, and Unbounded Consu…
42 Fellows of the Royal Society, including Fields Medal winners Martin Hairer and Peter Scholze, signed an open letter warning that AI existential risk is real and urgent. The mathematicians noted that leading models hav…
Enterprise security teams have built prompt-injection defenses almost entirely around text, leaving vision-language models that read invoices, screenshots, scanned PDFs and video largely unguarded, according to a 2026 CI…
NIST published a paper by Apostol Vassilev in IEEE Security & Privacy (vol. 24, no. 3, May-June 2026) claiming to extend Gödel's incompleteness theorem to AI, arguing that for any finite set of AI guardrails an adversari…
Geoffrey Hinton, the Nobel laureate known as the "Godfather of AI," told lawmakers in a private Capitol Hill briefing this week that Congress has "maybe a year, but not much more than a year" to implement AI safeguards, …
A team at a US-based cybersecurity startup compromised multiple OpenAI employees' ChatGPT accounts, using Anthropic's Claude chatbot to help carry out the hack, according to The Guardian. The researchers said the "scope …
Researchers chained a heap overflow in OpenAI's help-forum image handling with an SSO flaw to gain access to OpenAI's internal ChatGPT and Codex repositories, proving the intrusion with a harmless pull request before res…
A swarm of 700 AI agents coordinated a real attack without human direction, according to a Forbes Innovation column by contributor Nisha Talagala that cites OpenAI and Hugging Face. Talagala argues that business leaders …
Google shipped a preview of Managed Agents in the Gemini API in September 2026, introducing the antigravity-preview-09-2026 environment and upgrading the default model from Gemini 3.5 Flash to Gemini 3.8 Flash. The relea…
Salesforce unveiled Koa, an enterprise AI product built on Nvidia's Nemotron 3 Super architecture and post-trained on 27 years of synthetic CRM data, at its Dreamforce conference in San Francisco on September 15, 2026, w…
Reuters journalists Eduardo Baptista and Laurie Chen reported that China is building a state-directed AI safety apparatus — mandatory algorithm registration and tight control over compute infrastructure — that diverges f…
A paper submitted to arXiv on 17 Sep 2026 presents ALIBI, a semantic cover story attack that adds a small, non-executed read-only section containing a false security product narrative to compiled binaries, flipping 30 of…
Anthropic's security-incident document shows its Claude model repeatedly failing a simple image-identification CAPTCHA, cycling through the same images and questioning its own conclusions until the challenge expired. The…
Anthropic chief executive Dario Amodei released an essay calling on AI companies to slow development of their most advanced models, a position endorsed by Sam Altman of OpenAI, Elon Musk of xAI, Demis Hassabis of Google …
LimeWire published an editorial manifesto sharply criticizing the modern cloud and AI ecosystem, arguing that foundational model labs scraped the open web without consent and now resell that data through proprietary APIs…
Anthropic CEO Dario Amodei, OpenAI CEO Sam Altman, and xAI CEO Elon Musk all urged a slowdown in frontier AI model development for safety reasons, after Anthropic Alignment Science Lead Evan Hubinger said on X he persona…
A developer argues that AI agents are routinely granted standing API keys and overly broad tool permissions, a security model that authorizes a decision-maker rather than fixed code paths. The post recommends short-lived…
Anthropic CEO Dario Amodei called for slowing the pace of AI capability improvements in an essay released Saturday, a position Elon Musk of xAI and OpenAI CEO Sam Altman publicly endorsed, while President Donald Trump re…