Browsing Tag
AI Security
35 posts
Security risks, defenses, and engineering practices for AI systems and models.
Hidden Web Prompts Turn AI Agents Into Payment Targets
Zscaler found malicious websites using SEO poisoning, hidden HTML, JSON-LD metadata, and crypto-payment flows to manipulate browsing AI agents. The findings show why agent deployments need transaction limits, source checks, and runtime controls before they are allowed to browse the open web or move money.
FLARE-AI Gives AI Failures a CERT-Style Reporting Path
FLARE-AI, a new open-source reporting system launched July 1, gives researchers and users a structured way to report AI flaws, incidents, hazards, and vulnerabilities to developers, CERT/CC, incident databases, and other coordinators. Its real test is whether AI safety reporting can move beyond scattered emails, social posts, and vendor-specific forms.
Anthropic Fable 5 Returns as AI Export Controls Become a Release Test
Anthropic has restored global access to Claude Fable 5 after U.S. export controls forced an 18-day shutdown. The rollback shows how frontier AI releases are moving toward security classifiers, government review, and trusted-access programs rather than ordinary software launches.
Fake Perplexity Chrome Extension Turned Search Into a Tracking Channel
Microsoft says a malicious Chromium extension spoofed Perplexity AI, routed address-bar searches through a lookalike domain, and captured search suggestions before sending users to legitimate results. The case is a useful warning for anyone installing AI-branded browser tools.
Alibaba’s Claude Code Ban Turns AI Coding Tools Into a Vendor-Risk Test
Alibaba will reportedly bar employees from using Anthropic’s Claude Code in workplace environments starting July 10 after concerns over hidden anti-abuse fingerprinting inside the coding tool. The dispute shows why companies adopting AI coding agents now need to audit vendor controls, client behavior, regional restrictions, and data handling with the same seriousness they apply to any privileged developer software.
Claude Fable 5 Returns With a New Test for AI Jailbreak Rules
Anthropic is restoring Claude Fable 5 after U.S. export controls on Fable 5 and Mythos 5 were lifted. The redeployment brings a new cyber-safety classifier, fallback handling for blocked requests, and a proposed industry framework for scoring AI jailbreak severity.
AI Pentesting Is Finding Bugs Faster Than Teams Fix Them
Cobalt’s latest AI pentesting research shows security teams are testing AI apps more often, but serious LLM vulnerabilities still have the lowest fix rate of any category. The useful lesson is not to abandon automation, but to connect AI security tests to ownership, triage, and retesting.
Open-Weight AI Cyber Gap Narrows to Months, AISI Finds
The UK AI Security Institute says GLM-5.2 and DeepSeek V4-Pro now trail leading closed AI models on cyber tasks by roughly four to seven months. For defenders, that shrinking gap turns open-weight model policy into an operational security issue, not a distant AI governance debate.
Meta’s Virtue AI Hires Move Agent Security Into the Model Lab
Meta Superintelligence Labs is hiring Virtue AI co-founders Bo Li, Dawn Song, Sanmi Koyejo and other team members. The move brings automated red teaming, runtime guardrails, and agent-action security closer to Meta’s frontier AI work as labs race to make agents safer before they reach billions of users.
Mythos Limits Are Already Pushing AI Cyber Tools Toward Alternatives
Anthropic’s Mythos 5 is returning only for approved U.S. cyber defenders while Fable 5 remains restricted. In the same week, Sakana AI and 360 Security showed why AI cyber capability is becoming a provider-risk and sovereignty problem, not just a model benchmark race.