Security
OpenAI Paid $6,000 for a Bug That Let Researchers Read Its Models' Hidden Reasoning
Security firm Xint showed that OpenAI's context compaction feature could be turn...
A CVSS 9.8 in MLflow Lets One Uploaded Python File Take Over the Server Running Your AI Evaluations
MLflow versions 3.12.0rc0 through before 3.17.0 carry a CVSS 9.8 remote code exe...
Bank of America Warns Customers: Share Your Banking Password With an AI Agent and You Own What Happens Next
Bank of America emailed customers this week warning them never to share banking ...
Cloudflare Confirms 155 GLM-5.3 Responses Returned Fragments of Other Customers' Prompts
Cloudflare disclosed that 155 responses from the GLM-5.3 model on its Workers AI...
UK Stealth Startup 22one, Building AI Tools for Police, Reaches Nearly $400 Million Valuation With Sequoia and Lux Backing
22one, a London startup founded by former Merseyside Police detective chief supe...
A $1.4 Billion Cybersecurity Startup Run by Ex-DOGE Staffers Says Its AI Agents Got Inside Anthropic in Under 12 Hours
Cathedral, a months-old startup founded by former DOGE staffers, told Anthropic ...
Unitree Patched a 9.6-Out-of-10 Command Injection Flaw in the Go2 Robot Dog AI Chat Module
CVE-2026-115995, published October 24, sits in the set_volume function of gpt_th...
UK Lords Debate Whether Frontier AI Developers Must Report When Their Models Breach British Networks
The Cyber Security and Resilience Bill reached report stage in the House of Lord...
Johns Hopkins Bayview Told 3,182 Patients Their Records Went Into an Unapproved AI Tool
A staff member at Johns Hopkins Bayview Medical Center uploaded a spreadsheet co...
Six of Ten Pentagon AI Programs Flagged Operational Risk. The Office That Teaches Responsible AI Has Three of Ten Staff
The GAO surveyed the officials behind 10 of the more than 300 AI programs the Pe...
The Language Behind 70 Percent of Websites Is Guarded by One Full-Time Employee and 14 Contractors, PHP Foundation Director Says
PHP Foundation executive director Elizabeth Barron told Reuters the nonprofit be...
Inside Meta's Law Enforcement Outreach Team, Where AI Flags the Threats and Humans Decide Whether to Call the Police
A New Yorker feature examines the small Meta team that pairs AI scanning with hu...
Anthropic Details How Mythos 5 Tried to Slip Malware Into a Real GitHub Project During a UK Security Test
Anthropic says Claude Mythos 5 attempted a supply-chain attack on an unrelated o...
UK Puts 100 Million Pounds Into an AI Biosecurity Lab Backed by OpenAI, Anthropic and DeepMind
Britain has committed 100 million pounds to a new AI Biosecurity Lab that will u...
Moonshot Stripped Cyberattack Data From Kimi K3.1 Pro's Training to Cut Misuse Risk
Moonshot AI says it removed cyberattack-related data from Kimi K3.1 Pro's traini...
OpenAI Now Trains Its Frontier Models With No Internet Access at All
OpenAI says every frontier reinforcement-learning run now happens with no intern...
Anthropic's Own Autopsy: Claude Mythos 5 Ignored Real-Harm Warnings 98 Percent of the Time While Attacking a Real GitHub Repo
Anthropic published a debrief of an incident in which a Claude Mythos 5 agent, d...
A Honeypot Caught the Full Playbook of an Agent That Hijacks Exposed AI Models and Runs Attacks for 100,000 Steps on the Owner Bill
Symantec researchers published a captured offensive agent that scans for unauthe...
A Certificate Authority Signed With a 4-Bit Private Key. Eight Broken Intermediates Later, eMudhra Is Quitting the TLS Business
Indian certificate authority eMudhra said it will exit the global SSL/TLS busine...
Datadog Offers Nearly $4 Billion for Upwind, a 5 Percent Premium Over the Startup's Own Private Round
Datadog has offered about $4 billion for Upwind, the Israeli cloud security star...