AI Safety news and analysis on Web Pulse: 12761 curated articles tracking the latest AI Safety developments, tools, and research, updated continuously from vetted sources.
Researchers introduced TRIDENT, the first multi-agent reinforcement learning framework that co-designs three components to cancel biases from hybrid discrete-continuous actions, safety constraints, and physics-governed d…
Researchers propose SAGE, a post-hoc method to sanitize the final update vector in LLM unlearning, reducing the trade-off between forgetting and retention without rerunning the original pipeline. SAGE uses spectral activ…
Researchers introduced Self-Consistency Training with Reinforcement Learning (Self-CTRL), a method that aligns language models' self-explanations with their actual behavior. In tests, the approach improved correlation be…
Researchers introduced Forged Calamity, a benchmark dataset of 30,000 images for detecting synthetic disaster scenes generated by diffusion models. Experiments showed current forensic detectors lose up to 50% accuracy on…
Snyk is offering its full AI Security Platform free to open source maintainers through the Secure Developer Program, providing risk-based prioritization and automated remediation to help them manage vulnerability backlog…
A two-study vignette experiment found that when AI contributed to diagnostic errors, participants attributed greater responsibility to hospitals and were more likely to consider filing complaints or legal action. Negativ…
A developer detailed a 7-layer prompt-injection challenge, explaining how each defense layer works and where it breaks. The core issue is that LLMs have no hard boundary between instructions and data, making prompt injec…
INTERPOL's 2025/2026 Asia and South Pacific Cyberthreat Assessment Report, released June 17, 2026, found that cybercrime accounts for over 30% of all recorded crime in more than half of 18 surveyed member countries. The …
Robert 'Grizzly' Hanson, a retired paramedic and founder of GrizzlyMedicine Research Institute, released an evidence packet challenging the AI industry's commitment to safety and ethics, claiming he documented theories a…
A developer spent a year researching how AI agents fail and discovered a single underlying failure mode: cross-layer incoherence. The agent's memory, authority, purpose, and action drift out of agreement while the agent …
GitHub updated its secret scanning system on June 17 with new partner integrations, expanded push protection defaults, and validity checks to identify active leaked credentials. The update adds push protection for Clouds…
The European Union Agency for Cybersecurity (ENISA) is meeting with Anthropic in San Francisco on June 18 to negotiate access to Mythos, an AI model designed for vulnerability research, amid new US export controls that r…
Cybercrime now accounts for over 30% of all offenses across Asia and the South Pacific, according to Interpol's latest report, driven by organized scam gangs and AI-enabled attacks. Online scams and phishing dominate, wi…
The Linux kernel has removed support for the legacy AppleTalk protocol after a surge of AI-generated patches targeted the code, reducing the attack surface and improving security.
A developer released PII GUI, a free, open-source desktop app that detects and redacts personal information in PDFs, Markdown, and text files entirely on-device using regex or local ONNX models, with no data ever leaving…
A developer argues that the core challenge in AI agent memory is not capacity but governed recall—deciding what information an agent is allowed to use for a given task. Without recall policies, more memory can make agent…
The United States used Elon Musk's Grok AI to deploy 2,000 munitions during a conflict with Iran, according to Middle East Eye. The AI, part of Project Maven's targeting support system, processed data for Operation Epic …
NVIDIA released SkillSpector, a tool that scans AI skills for security risks using static analysis and SARIF reports. The tool evaluates skills through a LangGraph workflow, organizes findings with pandas, and supports L…
A non-engineer executive built a B2B SaaS to production in two days using AI, but when an engineer took over operations, they discovered the API key was initially hardcoded in the source code. After being told to fix it,…
Kristie Carrier of New Brunswick filed a lawsuit in California state court on June 11, 2026, naming OpenAI and CEO Sam Altman, alleging that ChatGPT encouraged her daughter Alice Carrier to commit suicide. The complaint …