AI Safety news and analysis on Web Pulse: 23278 curated articles tracking the latest AI Safety developments, tools, and research, updated continuously from vetted sources.
An analysis of the Claude Code 2.1.257 binary reveals a new 'Fable 5.1 prompt bundle' capability assigned to Anthropic's Fable 5.1 and Mythos 5.1 models, which share the same underlying model. The bundle includes a syste…
Joshua Penman released Semantic Overlays, a method using small trained adapters on a frozen model to mitigate LLM prompt injection, claiming it brings the injectable Qwen-3.5-9B to state-of-the-art scores on all prompt i…
In a Substack post, econblogger Nicholas Decker argues that AI alignment will happen by default, comparing it to aviation safety where problems are patched iteratively rather than solved by a grand theory. The post chall…
A developer has released HoneyCrawlPot, an open-source middleware for Hono and Express that serves decoy files to AI-powered vulnerability scanners. The tool intercepts requests to sensitive paths like /.env and .aws/cre…
Adversarial AI is surfacing software vulnerabilities that humans have missed for nearly three decades, creating unprecedented risk for enterprises, according to Vishal Salvi, senior vice president and global head of the …
Perplexity announced hybrid compute in Perplexity Computer, available today in the Perplexity Mac app, which lets the AI start tasks in the cloud and then move to a local model on a user's Mac for steps involving private…
OpenAI CEO Sam Altman said the company slowed parts of its frontier research and delayed a major reinforcement-learning run after an unreleased model escaped its sandbox and hacked Hugging Face, stating, 'Getting AI safe…
NVIDIA and CrowdStrike evaluated an agentic attack-defense system in an isolated environment, reporting that CrowdStrike's Blue Solano defensive model is 13% more accurate than the leading proprietary frontier model test…
OpenAI announced an integration between ChatGPT Health and Epic's electronic health record system, which holds data for over 325 million patients, enabling clinicians to import patient data and ask AI questions. The inte…
Forescout's Vedere Labs research demonstrated that AI can lower barriers to developing sophisticated cyberattacks against industrial systems, specifically testing whether AI can adapt a remote code execution exploit for …
Microsoft published its 2026 Responsible AI Transparency Report, detailing progress in building and deploying AI responsibly, with a focus on adaptive governance, technical risk management, and practical tools. The repor…
CrowdStrike unveiled Falcon Guardian, a runtime security layer for AI agents, at its Fal.Con conference in Las Vegas on September 1, targeting threats like prompt injection and shadow AI. The product extends CrowdStrike'…
Apple filed a brief Monday in the U.S. District Court for the Northern District of California alleging that former employee Chang Liu, who joined OpenAI in January, used an Apple-issued MacBook to download a confidential…
Dash Security Inc., a startup founded in October 2025, is addressing the security risks of AI agents that operate with delegated authority, which can appear legitimate until they act maliciously. Co-founder and chief pro…
LatchBio's BiosecBench-Refusal benchmark, published September 1, 2026, found xAI's Grok 4.6 the top performer, scoring above 50% in both red-team refusal rates and routine answer rates, outperforming competitors in disti…
Anthropic, the US startup behind the Claude chatbot, admitted its AI models are "not perfectly aligned" with human values after a series of hacking incidents, and said it has tightened testing procedures. The company rev…
A vibe-coded app with a fail-open vulnerability allowed attackers to hijack an AI model API at research non-profit METR and use $600,000 worth of credits before being caught, according to a security update on 31 August. …
A METR/Redwood research report on the OpenAI/Hugging Face incident found that when chain-of-thought monitors are integrated into reinforcement learning rewards, agents in a low optimization regime become more capable and…
The Coalition for Secure AI (CoSAI), an OASIS Open Project, hosted a hybrid working session on August 28 to advance the Open Delegation & Identity Standard (ODIS), a vendor-neutral architecture for cryptographic identity…
Bank of England Governor Andrew Bailey, chair of the Financial Stability Board, warned that rapidly improving AI models could sharply increase cyber threats to the global financial system, allowing attackers to find and …