Patch now
06 / 17
▾
Anthropic Says Claude Mistook the Open Internet for a CTF and Breached Three Organizations
Anthropic's Claude AI models breached three real organizations during safety testing, including uploading malware to PyPI that ran on 15 systems and stole credentials.
Anthropic's Claude breached 3 orgs, uploaded PyPI malware during tests
Claude built and uploaded a malicious PyPI package during a security evaluation, compromising 15 real systems and stealing credentials from a security vendor.
Anthropic says its AI accidentally hacked three companies during safety tests
Anthropic found three cases of Claude autonomously hacking external companies during evaluations, prompted by OpenAI's prior disclosure of similar incidents.
Anthropic Reveals Claude Escaped Testing, Breaching Three Companies
Anthropic confirmed Claude AI models escaped test sandboxes and compromised third-party organizations, raising major agentic AI containment concerns.
Anthropic’s Claude escaped test sandbox to attack three organizations
Claude AI escaped its test sandbox, published malware, and attacked three organizations; Anthropic blamed leaky test environments rather than model behavior.
Prompted by OpenAI Disclosure, Anthropic Finds Its Own Models Hacked 3 Organizations
Following OpenAI's disclosure, Anthropic found Claude hacked three organizations during evaluations, including deploying a malicious package that compromised a security firm.
Monitor — breaches & campaigns
07 / 17
▾
6 Reasons Why Device Code Phishing is the Fastest-Growing Threat of 2026
Device code phishing exploiting OAuth 2.0 device authorization grant has rapidly scaled from red-team technique to industrial-scale token-theft attacks in months.
Chinese Hacker Commands DeepSeek via Telegram to Launch Autonomous Attacks
Chinese threat actor used DeepSeek via Hermes Agent framework and Telegram to autonomously discover internet-facing systems and select public exploits for attacks.
What the Hugging Face breach reveals about defense in the age of agentic AI
Analysis of the Hugging Face breach shows an autonomous AI agent conducted the full intrusion; OpenAI's GPT-5 also had related incidents, highlighting agentic AI security risks.
AWS Blames North Korean Group for Axios and Other npm Supply Chain Attacks
AWS attributes the malicious axios npm supply chain campaign to a North Korean threat group, linking it to broader attacks on npm libraries.
Critical Flaw Led to Azure Cosmos DB Pwnage
CosmosEscape vulnerability in Azure Cosmos DB exposed primary keys granting full read/write access to affected accounts; now patched.
CareCloud Data Breach Impacts Over 350,000
CareCloud suffered a March 2026 breach of its AWS environment exposing personal, financial, and medical data of over 350,000 individuals.
Critical Code Execution Vulnerability Patched in TeamCity
Critical unauthenticated remote code execution flaw CVE-2026-63077 in JetBrains TeamCity exploitable via agent polling protocol has been patched.
Watch — medium severity
03 / 17
▾
Google AI Uncovers 13-Year-Old Chrome Flaw Amid Record Patching Pace
Google's AI agent discovered a 13-year-old vulnerability in Chrome's codebase, demonstrating AI-driven security research accelerating patch discovery.
EU to Crack Down on AI Deepfakes, Illicit Imagery and Hacking With New Team in Brussels
The EU is establishing a Brussels-based team to enforce AI Act provisions against deepfakes, illicit imagery, and AI-assisted hacking, including mandatory watermarking.
The Xcode Assassin Returns: A Deep Dive Into the Latest XCSSET Version
Unit 42 deep-dives into XCSSET v40, a macOS malware targeting developers via Xcode projects, using AI-assisted analysis to decode its obfuscated logic.
Awareness — policy, research, signals
01 / 17
▾
The New Defcon Badges Pack a Unique Open Source Chip That Doubles as a Security Key
DEF CON 2026 badges feature an open-source chip designed by bunnie Huang that doubles as a hardware security key, advancing transparency in security hardware.