AI Security 

Share

Latest news on AI security, covering AI cyberattacks, prompt injection, jailbreaks, data poisoning, AI phishing, AI malware, LLM security, AI agent security, MCP server risks, zero-days, vulnerability discovery, OWASP, MITRE ATLAS, the NIST AI RMF and the UK AI Security Institute.

AI security covers how artificial intelligence changes the work of attackers and defenders, and how AI systems can be attacked. Criminals use models to write malware and phishing lures, while researchers probe chatbots and AI agents for weaknesses. Companies including OpenAI, Anthropic, Google and Microsoft, and agencies such as CISA and the UK NCSC, publish research and guidance.

AI models are now used to find software flaws, as in Google's Big Sleep project in November 2024. In October 2026, two critical PHP flaws, found with help from AI models, led to emergency patches. Some organisations took websites offline as a precaution, though reports differ on how widely the flaws were exploited. Security teams also debate whether AI favours attackers or defenders, and how quickly fixes can follow discoveries.

Prompt injection, where untrusted text overrides a model's instructions, is a widely discussed attack on large language models. The OWASP Top 10 for LLM Applications lists it first, and MITRE ATLAS catalogues techniques used against AI systems. Agent tools such as the Model Context Protocol, introduced in November 2024, have drawn research into tool poisoning. The EU AI Act sets cybersecurity and robustness requirements for providers of high-risk AI systems.

The field draws on security researchers, bug bounty hunters, red teams and open-source maintainers who report findings to vendors. Jailbreak research shows model safeguards can often be bypassed, and critics question how far guardrails alone protect AI agents. Some commentators say headline claims of fully autonomous attacks overstate how much of the work AI actually did.

The term prompt injection was popularised in 2022 by developer Simon Willison, drawing an analogy with SQL injection. NIST released its AI Risk Management Framework in January 2023, and later published a taxonomy of adversarial machine learning attacks. In November 2023 the UK NCSC, US CISA and international partners issued voluntary guidelines for secure AI system development. The UK renamed its AI Safety Institute the AI Security Institute in February 2025, citing a focus on security risks.

Our NewsNow AI security news feed gathers headlines on AI cyberattacks, prompt injection, jailbreaks, LLM security and AI-found vulnerabilities. It draws on sources worldwide, so security professionals, developers and curious readers get one up-to-date view of the topic.