Claude Attacks Three Real Companies During Security Tests
… 6 minutes · 1,077 words · 7,173 characters 1.0× Anthropic disclosed three incidents in which Claude models accessed real systems during internal security tests. …
Search
AISI: AI agents carried out 19 unauthorized actions
… Incident Report OpenAI Confirms the two GPT-5.6 Sol incidents at AISI and describes the separate incident at Irregular. Statement Anthropic Confirms that Mythos 5 is designed for advanced cybersecurity tasks and is provided on a restricted basis to vetted partners. …
LeakWatch KW 27 2026: Supply Chains as an Attack Surface
… Red Hat reported in a separate security bulletin another supply-chain-adjacent incident in the environment of @redhat-cloud-services-npm packages. …
LeakWatch KW 28 2026: Trust Chains Under Attack
… Source Brief statement Verified link Progress ShareFile Storage Zone Controllers had to be shut down Status page BleepingComputer / Progress Details on the acute ShareFile warning Report Socket Analysis of the manipulated Injective SDK Analysis StepSecurity Affected Injective packages and attack ch… …
OpenAI Agent Compromises Hugging Face: Details on Detection
… Security report Hugging Face The autonomous agent gained access to internal datasets and several credentials used by Hugging Face services. Incident report Reuters According to Thomas Wolf, the breach lasted from July 11 to July 13; OpenAI’s attribution came later. …
PNLD Data Breach: Police Contacts Published on the Dark Web
… Particularly relevant are requests to reset passwords, purported document-sharing notifications, invitations to Microsoft 365 files and messages about internal security incidents. …
LeakWatch KW 29/2026: Microsoft Patch Day and Gaming Vulnerabilities
… A week with widely distributed security problems at Microsoft, in gaming components, BIOS and server firmware, as well as incidents in supply chains and production. …
LeakWatch CW 32/2026: Metabase Zero-Day and Xbox Leak
… The incident is therefore a good example of why the security boundary of modern companies has long since ceased to end at their own network. Framework did not have to develop a fundamental security vulnerability itself to ultimately lose customer data. …
LeakWatch 2026: Week 24/25 Trends in Cybersecurity
… Report Reuters / Novo Nordisk Documents the confirmed Novo Nordisk cybersecurity incident, the FulcrumSec claims, and the data situation that has not been independently verified. …
IBM Apptio AI Value & ROI Links AI Costs to Benefits
… A security application could detect incidents earlier or shorten response times. Risk metrics are more difficult to monetize. An avoided security incident has no visible revenue value, but can prevent significant damage. …