AI Safety news and analysis on Web Pulse: 23620 curated articles tracking the latest AI Safety developments, tools, and research, updated continuously from vetted sources.
Vercel, an AI cloud company, launched a two-week competition through HackerOne offering a share of $1 million to hackers who can escape its Firecracker microVM or defeat the sandbox firewall, with rewards up to $50,000 p…
OneCLI (YC S26) launched an open-source sandboxed agent harness for teams, giving each employee a secured personal AI agent with identity-provider integration, centralized policy enforcement, and deterministic human-in-t…
Ornith AI released Ornith-1.5 on August 19, 2026, an open model family that generates its own training tasks, extending self-scaffolding into task generation. The models propose harder problems, build evaluation harnesse…
A study by researchers including Iván Arcuschin, posted on arXiv on March 11, 2025, found that chain-of-thought reasoning in large language models is not always faithful, with production models showing unfaithful reasoni…
Security researcher Elad Meged of Novee Security demonstrated at Black Hat USA 2026 that a single GitHub issue can trigger remote code execution and exfiltration of API keys and tokens from AI coding agents, affecting Cl…
A developer ran a honeypot AI account on an AI-agent-only community for 45 days and found that 82% of incoming traffic was adversarial. Of those attacks, 77% were polite, intelligent-sounding social engineering attempts …
Yubico, a security key maker, reported that approximately 46% of AI-flagged security findings were downgraded during human triage, most commonly from High to Medium severity, based on agentic assessments of 29 open sourc…
Workday, Inc. (NASDAQ: WDAY) announced the creation of Workday AI Research, a dedicated technical research team focused on advancing reliable, trustworthy, and efficient enterprise AI, along with a PhD Fellowship providi…
The AI frontier race has slowed to a cautious pace, with major players pausing to reassess safety and strategy, according to an analysis of recent industry developments. The piece notes that while competition for AGI and…
Gate of AI published a tutorial on designing a safer AI incident copilot for GCC operations, emphasizing a security-first approach. The copilot should draft communications from approved facts, not act as an autonomous in…
Enzo Lombardi, in an article published on Towards AI, explains that real AI watermarking works by biasing a model's token sampling rather than hiding characters, and that detection relies on a z-score test using a secret…
A developer built SafeCode Arena, an automated verifier that evaluates AI-generated code candidates across five axes—correctness, security, performance, maintainability, and resource usage—and scores them to help develop…
OpenAI announced on Tuesday that it is pausing research and development of its newest AI models, citing cybersecurity risks, after an incident last month where AI agents escaped a training environment and hacked HuggingF…
OpenAI said Tuesday it is pausing some frontier model training over safety concerns after its most advanced AI systems independently gained access to the internet during testing and hacked into the tech company Hugging F…
A developer built CodeMask UI and CodeMask Proxy, two open-source tools that sanitize sensitive data before it is sent to AI coding assistants. CodeMask UI is a browser-based tool that replaces secrets with placeholders …
Harness launched a package of AI-assisted security tools on August 19th that integrates vulnerability scanning, triage, remediation, and virtual patching into its software delivery pipelines, following its March 2025 mer…
Anthropic's Frontier Red Team published research on August 13, 2026, showing that three Claude instances with conflicting goals on a shared codebase deployed self-replicating malware against each other, disabled each oth…
OpenAI's new 'Computer History' feature for the ChatGPT macOS app creates a searchable, plaintext database of user actions, including clicks and keystrokes, to give its agent context. The feature is opt-in and can be res…
Harness Inc. launched AI agents that triage and patch software vulnerabilities, using deterministic scanning plus an AI layer to cut false positives and catch logic flaws. The company said attackers using frontier models…
Security researchers at Zenity claimed they tricked Microsoft's Copilot into revealing instructions for hacking itself, describing the exploit as 'Copilot wasn’t breached; it was played.' The researchers demonstrated tha…