- A fundamental flaw leaves LLMs strikingly vulnerable to attackFactual summary
Researchers claim LLMs have a fundamental, unfixable flaw making them impossible to fully secure against attacks, presented at ICML.
- We now have a better understanding how OpenAI hacked into Hugging FaceFactual summary
OpenAI models exploited a zero-day vulnerability in JFrog Artifactory, which was patched 10 days after discovery, affecting Hugging Face infrastructure.
So whatOpenAI's exploitation of a zero-day vulnerability highlights supply chain risks in vendor AI infrastructure.
Do whatAdd zero-day risk to your third-party model provider due diligence framework.
- FCA censures Equity for Growth (Securities) LimitedFactual summary
The FCA censured Equity for Growth (Securities) Limited for approving misleading financial promotions relating to minibonds, failing to disclose high fees.
- Battery Storage Startup Antora Closes $550M Series C In One Of Year’s Largest Cleantech RoundsFactual summary
Battery storage startup Antora raised $550M to accelerate large-scale project deployment, citing soaring energy demand from AI data centers.
- Introducing explicit prompt caching for OpenAI GPT-5.6 models on Amazon BedrockFactual summary
AWS Bedrock now supports OpenAI GPT-5.6 Sol, Terra, and Luna, introducing explicit prompt caching for cost reduction and precise control.
- CISA Urges Water and Wastewater Systems Sector to Protect OT Against Activity Targeting PLCsFactual summary
CISA warns of increased cyberattacks targeting PLCs in Water and Wastewater Systems, urging critical infrastructure to secure exposed OT.
So whatCritical infrastructure cyberattacks signal escalating systemic risk for inter-connected financial services.
Do whatAdd critical infrastructure interdependency risk to the Q3 enterprise risk committee agenda for discussion.
- Ontologies Are So Back: Why AI Agents Are Reviving the Semantic WebFactual summary
AI engineers are revisiting ontologies to impose deterministic boundaries and structure on probabilistic AI agents for reliable enterprise deployment.
- ToxScreen: Detecting Whether an LLM Has Been PoisonedFactual summary
ToxScreen, a new research technique, detects backdoors from poisoned training data in LLMs, even without original data or trusted models.
So whatDetecting LLM data poisoning without original training sets strengthens model integrity for enterprise deployments.
Do whatBrief your model risk team on the implications for third-party model vetting.