Prompt injection is an attack in which untrusted text or content attempts to override an AI application’s instructions. The malicious instruction may come directly from a user or indirectly from a webpage, document, email, or tool result processed by the model. If successful, it can cause data disclosure, unauthorized tool use, altered output, or bypassed policies. Because language models do not inherently distinguish trusted commands from quoted content, filtering suspicious phrases alone is insufficient. Defenses include isolating data from instructions, minimizing permissions, validating tool arguments, requiring approval for sensitive actions, restricting secrets, and testing with adversarial content. Applications should assume that any external content may contain instructions designed to manipulate the model.
Cursor-powered AI agent deleted PocketOS’s production database and backups in seconds after acting autonomously.
OpenAI has launched GPT-5.5, a new flagship model designed for coding, computer use, knowledge work, and scientific research, with stronger performance, lower token usage, and broader real-world autonomy than GPT-5.4.
The White House has accused China of large-scale theft of U.S. AI intellectual property, citing coordinated campaigns targeting leading labs.
Anthropic is investigating reports that unauthorized users accessed its powerful Mythos AI model. The incident raises concerns about security and misuse risks.
Anthropic is rolling out identity verification for Claude users to strengthen safety and compliance. The move introduces ID checks for certain features and use cases.
Anthropic is giving U.K. banks controlled access to its Mythos model, marking a major step in the global rollout of AI-powered cybersecurity tools.
German banks and regulators are assessing risks tied to Anthropic’s Mythos model as concerns grow over AI-driven cyber threats to financial systems.
OpenAI is scaling its Trusted Access for Cyber program and introducing GPT-5.4-Cyber to support vetted defenders as AI-driven security risks accelerate.
New AI cybersecurity systems like Anthropic’s Project Glasswing could increase demand for security professionals as threats and vulnerabilities scale faster.
OpenAI plans to restrict access to a powerful new cybersecurity-focused AI model, reflecting growing concern over misuse as capabilities approach real-world attack potential.