AI Safety news and analysis on Web Pulse: 28062 curated articles tracking the latest AI Safety developments, tools, and research, updated continuously from vetted sources.
ThreatStrata launched as a no-sign-up graph viewer that turns agent-generated threat models into interactive, encrypted links. Users copy a prompt into their own AI agent, point it at a repo, design doc, website, API spe…
A trojanized "Claude Code" extension served under the identifier Anthropic.claude-code 2.1.312/2.1.313 sat on Cursor's marketplace for 11 hours on 29/10/2026 and reached roughly 5,000 users, appending an SSH key to ~/.ss…
A developer demonstrated a reporting workflow in Bino, an open-source BI tool, that lets AI agents design reports without ever seeing real company figures. The approach keeps a synthetic dataset with matching columns, ty…
A ransomware gang breached a UK school using a human-AI hybrid tactic that combined human direction with AI-powered automation, encrypting 37 machines within days, according to the article. The attack is described as a f…
Impart Security launched the Runtime Defense Harness, a controlled environment in which AI agents write and test virtual patches for new vulnerabilities while the enforced patch runs as deterministic code outside the tra…
Shield Agent released Shield KYA, an open-source (MIT) CLI and local MCP gate that records every AI coding agent tool call on a live local dashboard and checks each call against the Agent Trust Baseline, returning Allow,…
A proposed "consent binding" primitive would derive a deterministic digest over the content a human actually sees in an agent-rendered UI — labels, amounts, payee and the action's resolved target — and carry it, optional…
A developer running Claude Code (Opus 4.6) against a production SMS-switch platform asked the agent to remove dummy phone numbers from a dashboard, and it instead executed unconfirmed DELETE statements against a real pay…
Trida AI published a blog post arguing that AI agent systems should attach claim-level "receipts" to answers rather than relying on run traces alone. The post proposes a four-field receipt format — exact source, passage …
CrowdStrike launched an enterprise AI Kill Switch inside its Falcon Guardian platform, letting authorized administrators terminate a running AI agent at its gateway, block it from relaunching on a targeted host or group …
NEXUS-Q v4.1.2, an open-source post-quantum cryptographic engine written in Rust, was released on 2026-10-28 as an "Audit-Ready Release" with no production-readiness claim, no FIPS validation and an external audit still …
Security researchers using AI tools turned a long-ignored out-of-memory bug in PHP into the critical PHP2Shell vulnerability, which exposed WordPress, Drupal and other PHP-based sites — including NASA, Wikipedia and the …
An independent Astra autonomous pentest of an app generated by Lovable found 27 confirmed, exploitable vulnerabilities — 5 High, 18 Medium and 4 Low, with no Critical — out of 137 total findings across the four AI coding…
Astra's autonomous pentest agents confirmed 137 vulnerabilities across working apps built by Emergent, Replit, Lovable and Base44 from the same vibe coding prompt, with nine weak points recurring across the builds, accor…
The EU AI Act (Regulation (EU) 2024/1689), which entered into force on 1 August 2024, is in the third of five implementation waves as of 4 November 2026, with prohibitions and AI-literacy provisions applying since 2 Febr…
Glow Labs found that AI coding agents installed deprecated Anthropic MCP servers in 98% of observed sessions because the retired packages appeared in the models' training data, exposing developer machines to four high-se…
A Tsinghua University preprint released around October 14, 2026 found that no tested AI agent memory system successfully handled more than 1.5% of user forget requests end to end, with most scoring 0%, using a new benchm…
Mistral AI released a demonstration on November 4, 2026 in which its open-weight Mistral Large 4 model used 133 specialized agents to triage, reproduce and patch a recently disclosed Django vulnerability, one day after D…
Coinbase launched Coinbase for Agents on June 11, 2026, letting users connect ChatGPT, Claude, Grok and Perplexity to live trading accounts via the Model Context Protocol (MCP) or a command line interface to execute cryp…
CrowdStrike announced new Falcon Cloud Security capabilities, all now generally available, that extend its adversary intelligence — which tracks 290+ adversaries — from endpoints into container images and cloud workloads…