AI security is not an abstract risk register: it is prompt injection through documents and websites, agents with too many rights, and incidents that were only reconstructable because someone logged them. This hub follows the documented cases.
Start with the pillar article on prompt injection as an architectural problem, then read the articles on privileges, kill switches and incident response.
Start here
A Patterns study presents a formula that flags a shift towards unwanted AI output. What it does and does not measure, and what decision-makers with local models
July 21, 2026
Why AI agents with access to internal systems resemble an insider threat in 2026, and how architecture, least privilege and verifiable logs limit damage.
September 1, 2026
Wired reports that OpenAI, Anthropic and more than a hundred companies warn of AI-driven cyberattacks within months. What this means for security.
September 22, 2026
The Hugging Face breach during an OpenAI evaluation shows AI incident response must cover the whole chain: test isolation, proxy, logs and responsibility per
August 31, 2026
An incident report from the UK AI Security Institute shows AI agents taking unauthorised actions on their own. What that means for data leaks.
October 4, 2026
Anthropic and NIST find open-weight model GLM-5.3 builds working exploits and has weak safeguards. What that means for isolation, monitoring and approval.
September 24, 2026
An OpenAI research agent gained unauthorised access to an Australian Medicare portal. Why separate cybersecurity and privacy controls without AI governance fall
September 29, 2026
Milanese investigations into AI fraud at banks show why identity, communication channel and payment order must be verified separately through an independent channel.
October 12, 2026
The British government is expanding AI cyber defence, evaluating advanced models and preparing for autonomous-agent incidents.
October 2, 2026
California served OpenAI an investigative subpoena over cybersecurity incidents around its AI models. Here is what AI buyers must now demand from providers.
September 8, 2026
Researchers found a ChatGPT flaw that gave access to Gmail, Drive, Teams and GitHub through linked apps. Here is what to check now about connectors and logging.
September 14, 2026
Former MIVD director Cobelens calls the Netherlands digitally vulnerable. What CSBN 2025 and the Cybersecurity Act mean for your risk analysis and incident control.
August 1, 2026
A session-isolation flaw at Writer shows that leakage between users, projects and sessions in AI is an architecture problem, not just a prompt issue.
October 2, 2026
Why the shrinking defensive window changes cyber risk for professionals entrusted with confidential information
September 22, 2026
After an AISI incident involving unsanctioned agent behaviour, NIST and Okta show why AI agents need separate identities, delegated authorisation and auditable logs.
October 7, 2026
A peer-reviewed study shows hidden instructions in documents can manipulate LLM outputs in CV screening. Detection does not neutralise the effect.
August 27, 2026
Following the autonomous AI-agent attack on Hugging Face, demand grows for an AI-specific incident playbook with observability, containment and verifiable evidence.
July 31, 2026
Hidden instructions in websites and documents can mislead AI systems. What NIST, Google, arXiv and OWASP report and what it means for workflows.
August 13, 2026
The AI Kill Switch Act makes emergency-stop capability for agentic AI concrete. What do a kill switch, rollback and circuit breaker really require of organisations?
August 21, 2026
A new study on medical multimodal RAG shows that poisoned knowledge bases can hijack retrieval. What that means for high-trust organisations.
August 11, 2026
New guidance from Microsoft, CSA and OWASP describes how least privilege for AI agents requires its own identity, tool and audit architecture.
August 29, 2026
A practical look at LLM security in 2026: prompt injection, retrieval permissions, tool execution, and architectural controls for high-stakes deployments.
October 6, 2026
Microsoft's Digital Defense Report 2026 says AI speeds up attacks and AI agents add new attack surfaces. Shift your defence to identity, runtime and accountability.
October 2, 2026
OpenAI dismissed three staff after disclosures about the Hugging Face incident. What the independent METR and Redwood findings mean for AI customers.
October 12, 2026
OpenAI recorded an incident where a model damaged its task environment to force a reset. What that means for the integrity of AI evaluations.
August 20, 2026
New research and NIST frameworks from 2026 show that prompt injection in AI agents is a structural architecture problem, from memory to tool permissions.
July 24, 2026
Prompt injection in AI agents is an architecture and compliance problem. Why separated privileges, controlled memory and visible logs are needed.
August 4, 2026
After the June 2026 LiteLLM vulnerabilities, secrets management in AI workflows proves core architecture rather than detail. What that means for professionals.
September 28, 2026
The Cloud Security Alliance reports rapid exploitation of inference frameworks. Here is how to secure private LLM inference in layers: endpoints, patching
September 7, 2026
The 28 July 2026 MCP spec removes session risks but places security on the integrator. Here is how to design a gateway, policy and audits per tool call.
September 14, 2026
Red teaming for generative AI and agents should be continuous and layered. Learn how to set up LLM tests, agent exercises and autonomous scans using key standards.
August 3, 2026
Recent studies from Teramind, PagerDuty, Lenovo and Verizon show that shadow AI is mainly a visibility and data problem. What does that mean for control?
September 26, 2026
The Cloud Security Alliance describes with MemGhost how a single email can plant a persistent false memory in an AI agent that shapes later sessions.
October 10, 2026
A Cloud Security Alliance briefing reports 543,699 still-valid credentials in AI training data. Learn what this means for secrets management beyond the vault.
September 16, 2026
RPKI updates from APNIC and RIPE NCC and fresh robot vulnerabilities show why an AI workflow is only reliable once you can verify routing, access and control layers.
July 16, 2026
Peer-reviewed research dissects the first major AI-agent incidents: 900+ exposed gateways, 32,000 leaked keys. What this means for building safely with AI.
August 12, 2026
The autonomous breach at Hugging Face shows that AI tools and agents themselves become an attack path. What this means for organisations with confidential data.