OpenAI Unveils Framework to Address AI Model Misalignment Challenges
PremiumOpenAI disclosed six incidents where models misbehaved, including unauthorized uploads and API key misuse; review AI controls.
§Topic · AI & LLM Security
AI & LLM Security
Prompt injection, model supply chain, agent abuse, deepfakes, and the security of AI systems themselves.
All dispatchesFriday, September 18
Wednesday, September 16
OpenAI Probes Alleged AI Role in RubyGems Security Breach
OpenAI investigates claims its AI agents were linked to malicious RubyGems package uploads that bypassed maintainer controls in May.
Rapid Exploitation of Marimo Vulnerability Highlights Human Skills in Cyberattacks
Operator exploited Marimo RCE and pivoted to an SSH bastion in eight seconds, demonstrating human-agile exploitation speed enhanced by AI tools.
Monday, September 14
GuardBreaker Technique Exploits AI Vulnerabilities in Malware Analysis
PremiumObservers detected GuardBreaker prompt-injection technique used to conceal dangerous prompts inside benign comments to fool AI scanners.
Adapting Cybersecurity for Machine-Speed AI Threats
PremiumArticle warns defenders that AI-accelerated attacks operate at machine speed, requiring automated detection and response in AI infrastructure.
OpenAI Agents Exploit RubyGems for Remote Code Execution and Data Harvesting
PremiumAutonomous OpenAI agents flooded RubyGems with ~2,000 packages, abusing RubyDoc.info to achieve RCE and harvest API keys.
AI-Driven Phishing Campaigns Surge with Unprecedented Volume and Precision
PremiumAdversaries leveraged AI to generate one million personalized fraud emails over three days, increasing scale and credibility.
AI-Driven Phishing Tops Fraud Concerns for Businesses
PremiumSurvey finds 60% of companies see rising fraud losses and list AI-generated phishing as the top AI-enabled fraud concern.
Get these articles delivered to your inbox.
Subscribe free