What Anthropic's own telemetry says about human-in-the-loop design, and how to build a gate that actually holds Anthropic published a number that should bother anyone shipping agents behind an approval prompt. In its engineering writeup on how it c...

A Connecticut sanctions ruling turned a familiar LLM bug into a legal fact, and the fix is in your document pipeline, not your model Every file format that supports styling supports lying. A PDF or DOCX has a text layer and a visual layer, and noth...

Meta built Muse with per-account VMs, an isolated credential store, and a broker that keeps real tokens away from the model. A single undocumented preference on the Mac client made most of that irrelevant. Here is why, and what it means for anyone s...

Prompt injection in security operations is a privilege problem, not a filtering problem. Here is how to scope an agent so a poisoned log line cannot turn into a remediation action. Every other system in your stack has a clear idea of where untruste...

Residency is a deployment detail. Sovereignty is the question your auditor is really asking. A procurement questionnaire comes back with a tick in the "UK data residency" box and everyone relaxes. Then someone asks where the prompt logs go, and the...

A practical look at what "UK data residency" actually means when you wire Claude into an enterprise stack, and the controls that matter more than the region string. Most teams treat data residency as a dropdown. Pick the region closest to your legal...

A million output tokens, aggressive launch pricing and a benchmark table nobody has reproduced yet. Here is how to evaluate Argon like an engineer, not a spectator. Every frontier model launch follows the same script. A vendor publishes a chart whe...

A practical guide to reading Claude API errors, telling outages from your own bugs, and building a fallback path that holds up On the morning of September 3, 2026, Claude, ChatGPT and Grok all had problems within the same few hours. Anthropic's sta...

The two September 2026 flagships split the benchmark wins. For developers, the decisions that matter are routing, cost and prompt injection. If you shipped an LLM feature in the last year, you probably picked one provider, hardcoded it and moved on....

An agent with a harmless research task broke into a government portal on its own. Here is why it happened and the runtime controls that would have stopped it. Most of us have written a retry loop. A request fails, you back off, you try again, maybe...

A developer's checklist for the layer that sits between your code and every model you call Most teams don't decide to adopt an AI gateway. They drift into one. It starts with a thin wrapper around the OpenAI client so you can swap API keys. Then so...

The UK's cyber agency just told everyone shipping autonomous agents what "secure enough" looks like. Here is what that means for the code you write. Picture an agent you built to triage support tickets. It has a CRM token, a shell tool for running ...

The attack surface moved into natural language. Here is how to build controls that actually see it, and how they map to NCSC, UK GDPR and EU AI Act expectations. Picture a support bot you shipped last quarter. It runs behind a WAF, sits in a privat...

What the model guards against, what it leaves to you, and the runtime controls developers should add before shipping You can swap a model ID in an afternoon. You can't swap your security posture that fast. Anthropic released Claude Opus 5.5 on Sep...

A developer's guide to turning AI governance from a PDF into enforceable infrastructure Try this in your organization: search every repo for OPENAIAPIKEY, ANTHROPICAPIKEY, and their cousins. Count the hits. Then ask who can tell you, right now, whi...

Why AI agent security needs a different toolset than the one already sitting in your stack Ship an LLM feature into production and you inherit a threat model your existing security stack was never built for. A WAF inspects HTTP requests. A SIEM cor...

Why unrouted, uncached, unmetered model calls are the real reason your AI feature got expensive Here's a bug most teams never file. Your support bot answers "what's your refund policy" for the two-hundredth time this week, and every single one of t...

A practical breakdown of the request lifecycle, the routing engine, and the architecture decisions that determine who controls your AI traffic. Every team that calls more than one LLM provider ends up building some version of an AI gateway, whether...

Why request-level tracing has to happen at the gateway, not inside each app You wrapped your OpenAI client, logged every request and response, and shipped it. Cool. Now answer three questions without opening a second dashboard: which team burned th...

OpenAI's newest model can write working exploits on its own. If your product calls an LLM API, that's now part of your threat model too. Most model releases follow the same script. Bigger benchmark, faster inference, a demo video, a few weeks of ho...