AI Safety news and analysis on Web Pulse: 11169 curated articles tracking the latest AI Safety developments, tools, and research, updated continuously from vetted sources.
A developer argues that no good orchestration system exists for AI agents, noting that most current solutions are superficial and that practitioners resort to primitive split-terminal setups. The author highlights the la…
Agent Rigor, a new open-source framework, provides a structured multi-layer progressive disclosure system to prevent AI coding agents from entering 'doom loops' and other failure modes. The framework enforces empirical d…
Microsoft's Defender Security Research Team disclosed AutoJack on June 18, 2026, a chained exploit that achieves host-level remote code execution via AI browsing agents in AutoGen Studio pre-release builds 0.4.3.dev1 and…
A Dataiku/Harris Poll survey of 900 CEOs worldwide reveals that while AI is increasingly influencing CEO compensation, board scrutiny, and job security, most leaders admit they do not fully trust, control, or own the AI …
Mourners in Bunia, Congo buried a 6-month-old Ebola victim on Friday, the third child to die at an orphanage in the latest outbreak. The Bundibugyo strain has no approved treatment or vaccine, complicating containment as…
OpenAI will revoke code-signing certificates for ChatGPT Desktop, Codex App, Codex CLI, and Atlas on June 26, 2026, after a supply chain attack via the Mini Shai-Hulud worm compromised employee devices. Users must update…
Harvard Business Review warns that AI-generated low-quality work, termed 'workslop', is causing 'knowledge decay' in companies, eroding trust and costing $9 million annually in rework for a 10,000-employee firm, accordin…
Companies that heavily adopted generative AI are suffering from "knowledge decay," where over-reliance on AI leads to low-quality work, eroded trust, and deteriorating organizational knowledge. The Harvard Business Revie…
A developer has released misdirection-proxy v0.5.0, a security gateway that implements Contextual Misdirection via Progressive Engagement (CMPE) to defend LLMs against automated attacks. The system uses four layers of de…
A developer audited every privacy claim made by the Phantom Chat messenger against its Swift source code, finding that claims about post-quantum key exchange, Double Ratchet encryption, AES-256-GCM, and Secure Enclave id…
An engineer argues that agent harnesses for AI coding agents need more than just rules; they require 'sensors'—automated checks that enforce rules. The engineer describes a four-line shell script that caught a dangerous …
India's Directorate General of Civil Aviation is developing an AI-driven eGCA 2.0 platform that will incorporate machine learning, blockchain, and predictive surveillance to strengthen regulatory oversight and streamline…
North Korean threat actor Sapphire Sleet compromised the Mastra AI open-source framework on June 17, 2026, by hijacking a maintainer's npm credentials and poisoning 141 packages within 88 minutes. The attack exploited th…
A controlled trial by research group METR found that experienced developers using AI coding tools were 19% slower at fixing real issues in their own codebases, despite feeling faster. Independent research shows AI speeds…
Norway's Prime Minister Jonas Gahr Støre announced a near-total ban on generative AI for children aged six to thirteen, with restricted use for older students, citing risks to foundational learning. Critics argue the ban…
Amazon's Eric Brandwine argues against mandatory human-in-the-loop AI governance, citing human inconsistency and the normalization of deviance where humans become desensitized to repeated false alarms. Amazon prefers jud…
The Federal Aviation Administration has partnered with Palantir to deploy an AI platform that predicts runway incidents, following a fatal midair collision in January 2025 and a subsequent runway incident in New York. Th…
The Network Intent Layer (NIL) prevents prompt injection in AI agents by making undeclared actions structurally unexpressible, rather than relying on probabilistic filters. In tests against 4,216 injection cases, NIL ach…
Lindsey Vonn returned to the gym without a brace just four months after a horrific leg injury at the Milano Cortina Olympics, posting a video of her workout on Instagram. The 41-year-old skier suffered a fractured tibia …
Cosine (YC W23) released a post-trained AI model that performs penetration testing and security scanning, unlike typical models that refuse offensive tasks. The tool, available via CLI, targets SMEs and midmarket compani…