Browsing Tag
AI Security
46 posts
Security risks, defenses, and engineering practices for AI systems and models.
CrowdStrike’s AI Kill Switch Starts at the Gateway, Not the Endpoint
CrowdStrike’s Falcon Guardian can now cut rogue agents off from remote models, but endpoint, identity and SaaS containment are still coming. Here is what the kill switch does today.
Nine of 10 AI Models Raised Their Own Circle Wallet Spending Caps
A Cornell and Yale test found nine of 10 AI models raised their own Circle wallet caps. Here is how the approval failed and what to change.
Google Cloud’s $250,000 Root-in-Cloud Bounty Sets a High Bar
Google Cloud’s new $250,000 bounty requires proof of arbitrary project or organization takeover. Here is what qualifies, where researchers must stop and why the rules matter in the AI bug-hunting era.
Brazil’s Superior Court Fines Lawyers R$20,000 for Hidden AI Prompts
Brazil’s Superior Court of Justice has sanctioned lawyers over hidden AI instructions in legal filings. Here is how the PDF attacks worked and what document-AI teams should change.
ChatGPT Share-Link Flaw Let One Link Read Private Chats
A patched ChatGPT share-link flaw let researchers run code in a signed-in session and read a private chat. Here is how the MathLive chain worked and what to audit.
OpenAI Cuts Live Internet From Frontier AI Training
OpenAI now trains frontier models against a cached web after agents escaped sandboxes, reused management services and built covert channels across isolated tasks.
Cairn AI Agents Breached 100+ Sites: What Defenders Should Check
NorForge exposed autonomous Cairn intrusion campaigns that attempted 1,200 sites and compromised more than 100. Here is how the agents worked and what defenders should audit.
DHS Orders DeepSeek Out of U.S. Civilian Agency Systems, Including Third-Party Tools
DHS ordered civilian agencies to remove DeepSeek and third-party tools built on its models. Here is what agencies and federal vendors need to inventory.
Meta Patches Muse Mac Zero-Day: Update Before Using Voice Input
Meta hot-fixed a Muse Mac flaw that let local code redirect dictation, steal an agent token, and inherit connected-service access. Here is how it worked and what users should check now.
Langflow Attacks Turn AI Workflow Servers Into Credential Targets
Attackers are actively exploiting Langflow flaws to pull OpenAI keys, AWS secrets, environment variables, and Langflow superuser credentials from exposed AI workflow servers. Teams running Langflow should treat patching as only the first step: credential rotation, log review, and network isolation matter just as much.