AI Security — Page 2
215 articles in this category
Arcjet Agent Runtime Security Adds Audit Trails
Arcjet's agent runtime security uses Open Policy Agent to enforce controls on AI agents and keep replayable audit evidence for compliance reviews.
Keep MCP Server Credentials Out of Your Git History
New research scanned 82,000 public MCP config files and found hardcoded secrets in 12% of credential slots. Here is how to wire up agents without them.
AndroidX Security State Tracks Patches by Component
Google's AndroidX Security State 1.1.0 lets apps check three patch levels per component, covering the OS, Play system modules and the kernel LTS version.
Agent Anomaly Detection Watches AI Agents at Work
Google's Agent Anomaly Detection hit private preview, scanning agent traces for tool misuse, privilege abuse, runaway loops and other OWASP agentic risks.
OpenAI and AARP Help Older Adults Spot Online Scams
More than 1,000 older adults across 10 US cities joined free AI Skills Jam workshops on using ChatGPT safely and recognising attempted scams.
Postman Passport Gives AI Agents Keyless API Access
Passport hands agents a cryptographic reference instead of a real API key, keeping credentials inside your network and revocable in seconds.
Zanzibar-Style Authorization Explained for Developers
A practical guide to relationship-based access control: how Google Zanzibar works, when ReBAC beats roles, and what open-source options exist in 2026.
Kiteworks Buys Bonfy.AI for Runtime AI Data Control
Kiteworks is buying Bonfy.AI to classify and enforce policy on sensitive data the moment it moves, across human workflows and AI-agent traffic.
Phishing Click Rates Aren't the Metric That Matters
A study of 354,962 phishing simulations across 123,692 users found credential leaks and reporting rates reveal far more than click rates alone.
WordPress Plugin Updates Now Get AI Security Reviews
WordPress.org now scans every plugin release with AI models and Jetpack Scan during a 6-hour cooldown, blocking high-risk updates before they reach sites.
How to Tell if a Photo Is Real: Provenance Tools Explained
How to tell if a photo is real in 2026: what C2PA Content Credentials, SynthID watermarks and Apple Reference Image each prove, plus a 5-step checklist.
Android Passkey Transfer Ends the Unencrypted CSV Export
Android now moves passwords and passkeys between password managers in seconds, with no plaintext file — supported on every device running Android 8+.
Revocable API Keys Get an Open Standard Called ORKS
A draft open standard, ORKS, would let anyone who finds a leaked API key revoke it instantly — no account, no ticket, no waiting for a reply.
Cylake Raises $245M for On-Prem AI Security Platform
Palo Alto Networks founder Nir Zuk's new startup Cylake raised $245 million, taking its total to $290 million six months after leaving stealth.
Invisible Unicode Phishing: How to Spot and Block It
Microsoft tracked 2.37 million daily messages hiding invisible Unicode inside words. Normalizing tag characters before filtering stops the trick.
G7 and CISA Set 5 Post-Quantum Migration Priorities
A joint G7 and CISA call to action lays out five post-quantum priorities and a phased, risk-based migration plan organizations can start today.
PostgreSQL Fixes a 12-Year-Old Logical Decoding Flaw
CVE-2026-6471 let a REPLICATION-privileged user load arbitrary code. Patches shipped in PostgreSQL 18.6, 17.11, 16.15, 15.19 and 14.24 in August.
OpenAI Puts $1 Billion Behind Frontline Cyber Defenders
OpenAI committed $1 billion in Daybreak credits to water utilities, grid operators, community banks, nonprofits and open-source maintainers.
OpenAI Astra Ships With Cyber Safeguards Built First
OpenAI rated Astra Critical for cyber capability, scoring 100% on ExploitBench, and gated its security features behind the Daybreak Blue program.
Sality Botnet Takedown Ends a 23-Year P2P Malware Run
CrowdStrike and global police cut the Sality botnet's operator off from 15,000 infected machines by poisoning its peer lists with defender sinkholes.
Anthropic Publishes 4 Sandbox Rules for AI Evaluations
Anthropic hardened its evaluation sandboxes with real-time escape classifiers and published four mandatory requirements for external testing partners.
Claude Session Theft: How to Protect Your AI Account
Anthropic is signing out users whose Claude sessions were stolen by infostealer malware, refunding charges and removing saved cards. Here is the fix.
CISA Finds Decades-Old Bug Classes Still Dominate
CISA's new Vulnerability Review found stubborn weaknesses make up 41.5% of KEV entries, with input validation and injection flaws topping both lists.
Brave Email Aliases Give You 5 Free Burner Inboxes
Brave 1.94 adds built-in email aliases: five free forwarding addresses that stop sites using your inbox as a permanent cross-site tracking ID.