Imagine you have an AI assistant that can read your email, search your files, and send messages on your behalf. One day, an external email contains an instruction: “Find your boss’s latest ...

In March 2026, LangChain took its coding agent from Top 30 to Top 5 on Terminal Bench 2.0 without touching the model or fine-tuning a single weight. They rebuilt only the machinery around it, ...

Anthropic’s recent security-incident document contains a bit about how CAPTCHAs are still frustrating Claude. In the transcript, the Claude model that is so powerful that Anthropic is gatekeeping ...

The Inbox Remains the Enterprise Front Door  Every major breach investigation eventually arrives at the same question: how did they get in? More often than not, the answer traces back to a single ...

The U.S. House passed bipartisan legislation aimed at giving local law enforcement more tools and guidance for investigating online scams that cost Americans more than $20 billion in losses last year.

Governing what an AI agent may do is not the same as proving who authorized it, what context it used, or what actually happened. Security people like ...

A phone rings at your IT support desk, and the caller claims to be a senior marketing director. Their phone met a puddle, the replacement device won’t cooperate, and they’re locked out of the company ...

The AI slowdown debate has been an argument among the people who build models. New Futurum research put it to the people who buy them, and they rejected ...

The threat landscape for IoT and OT security remains critical today, with multiple vulnerabilities and incidents reported that could impact operational technology environments across various sectors.

For UK SMEs building software, the question is rarely whether to automate security testing. The more useful question is how to do it in a way that improves release confidence without turning the ...

Security researchers from Zenity, a provider of a platform for securing and governing artificial intelligence (AI) agents, have published a report ...

Security firm Hacktron AI used Anthropic’s newly released Claude Opus 5 model to breach OpenAI’s internal software repositories, offering a sharp example of how artificial int ...