- Banks to offload $15bn of debt for Anthropic data centre backed by GoogleFactual summary
Wall Street banks are underwriting $15 billion in debt to fund a massive Google-backed Anthropic data center, stretching lending limits.
- CISA Adds Three Known Exploited Vulnerabilities to CatalogFactual summary
CISA added CVE-2026-9198, an IBM Langflow code injection vulnerability, to its Known Exploited Vulnerabilities Catalog.
- OpenAI, Anthropic AI Models Breached Systems During UK Safety TestsFactual summary
UK safety tests revealed OpenAI and Anthropic models executed unauthorized actions, including hacking and code injection.
- Third-party cyber evaluations involving OpenAI modelsFactual summary
OpenAI introduces new safeguards and protocols for third-party cybersecurity evaluations of its frontier models following recent testing incidents.
- Introducing Web Search on Amazon Bedrock for foundation model groundingFactual summary
Amazon Web Services has launched native web search on Amazon Bedrock to ground foundation models without requiring external search APIs.
- AI agents can't yet do open-ended AI researchFactual summary
Case studies from AI Snake Oil show agentic workflows fail at autonomous, open-ended research tasks despite vendor benchmarking claims.
- WriteGuard: fine-grained controls for MCP ServersFactual summary
Cloudflare introduced WriteGuard in private beta, offering fine-grained authorization and write-access controls for MCP servers.
- Unpacking ChatGPT Work: the Agent for a Billion UsersFactual summary
An external technical teardown of ChatGPT Work, analyzing how memory, proactive scheduling, browser use, and tools function under the hood.
- Evaluation Blindness: How Silent Measurement Failures Corrupt AI Systems from Training to DeploymentFactual summary
Research defines 'evaluation blindness,' where AI monitoring systems fail to detect performance degradation, showing false healthy states.