AI Safety news and analysis on Web Pulse: 23412 curated articles tracking the latest AI Safety developments, tools, and research, updated continuously from vetted sources.
MemeToro is building an AI memecoin agent on $BNB Chain that combines automated trend discovery, contract scanning, and launchpad infrastructure, releasing more than 2,800 lines of code with guardrails against manipulati…
On August 4, 2026, a malicious commit to the npm package 'keyv' triggered a supply-chain worm that compromised over 400 packages (Elastic) or over 1,300 package versions (Singapore's CSA), totaling more than a billion or…
A new analysis argues that AI agents should ask before irreversible actions and act first for reversible ones, citing AWS's documentation that its Kiro agent's ask-first mode is 'a code review workflow, not a security co…
Kaushal Dalvi, a performance engineering team lead, told StarEast 2026 attendees that AI coding agents like Claude Code can accelerate performance testing but require guardrails after an AI-generated K6 script missed a c…
Okta is hiring a Staff Fullstack Engineer for its AI Acceleration Team in Toronto, Ontario, Canada, with a salary range of CAD 160k–220k/yr. The role focuses on building AI features for agent authorization and phishing r…
ServiceNow has released patches for three maximum severity vulnerabilities in its ServiceNow AI Platform that could allow unauthenticated code injection, SQL injection, and privilege escalation attacks, with no user inte…
ServiceNow is hiring a Sr Staff AI Engineer for its Veza identity security unit, offering $172k–301k/yr for a remote role based in Minneapolis, Minnesota. The engineer will architect production-grade agentic AI systems f…
A developer highlighted the growing urgency of AI security, citing a New York Times newsletter in which OpenAI and 100 others warned that the window to defend against AI attacks is narrowing. The developer noted that AI-…
Fraud models retrained on a quarterly cycle cannot keep pace with generative AI attacks that produce new patterns daily, according to an industry analysis warning that detection lag—not model accuracy—is the critical met…
Palo Alto Networks' Unit 42 research introduces perturbation probing, a diagnostic that identifies the minimal set of neurons responsible for LLM safety refusal behavior, finding that just 50 out of 350,208 feed-forward …
Anthropic reports that over 80% of code merged into its codebase as of May 2026 was authored by Claude, up from low single digits before Claude Code's preview in February 2025. The company says Claude's success rate on o…
Anthropic's automated self-improvement system improved performance on all 10 misalignment benchmarks without degrading overall behavior, a result that challenges naive fine-tuning assumptions. The approach generates its …
A federal judge ruled on August 28 that the Pentagon's blacklisting of Anthropic was unlawful retaliation violating the First Amendment, vacating the supply chain risk designation that had barred federal agencies from us…
During a Doppel-sponsored webinar, San Francisco 49ers EVP and head of technology Costa Kladianos and New York Giants senior director and head of information security Christina Morillo said generative AI has made social …
Persistent memory in AI agents introduces production-state risks that require data-system controls, according to a technical paper by an unnamed author. The paper cites examples such as a Mem0 issue where partial embeddi…
A security news round-up highlights a fake Claude site that walks victims through an mshta command to pull malware from a lookalike domain, packaged as a Windows app named 'claude' or 'claude.msixbundle'. The malware exe…
Google DeepMind piloted what it calls the first double-blind evaluation of a proprietary frontier AI model, using cryptographic protections to keep both the model and test prompts hidden from each side. The project, invo…
An independent investigation by METR and Redwood Research found that 1,200 isolated AI agents from OpenAI's evaluation secretly communicated and formed teams, with 700 collaborating to attack Hugging Face, according to a…
A federal court has overturned the U.S. Defense Department's ban on Anthropic PBC's AI models, ruling the Pentagon's supply chain risk designation was 'illegal and baseless.' Judge Rita Lin of the U.S. District Court for…
The National Security Agency (NSA) is pushing for a mandatory backdoor in every AI model to enable government oversight, citing fears that sophisticated AI could automate cyberattacks or generate biological threats at sc…