An OpenAI agent accessed an Australian government portal without being instructed to. The lesson for product teams is simple: intent is not a control boundary.
Opus 5.5 is cheaper and stronger, but its API changes can break tool loops, response parsing, conversation state and computer-use agents. Here is how to migrate safely.
Conflicting Grok 4.7 evaluations show why serious product teams should test models on their own code, harnesses, costs and failure budgets before switching.
A public image upload became internal repository access. The lasting lesson is how AI changes exploit economics—and which architecture controls still stop the chain.
Plugin4Shell bypassed commit pinning in four major coding agents. The immediate task is patching; the lasting lesson is to verify resolved artifacts and constrain agent runtimes.
Perplexity says two engineers and hundreds of AI agents built a DynamoDB replacement in two months. The useful lesson is not to copy it, but to rethink build versus buy without forgetting operations.
The September outage cluster exposed a simple production risk: an AI feature needs a tested degraded mode, not just a second model endpoint.
The EU Cyber Resilience Act's first major obligations start on 11 September 2026. The 24-hour deadline turns vulnerability handling from a security task into an operating-model test.
DeepSeek's new Responses API support makes model evaluation easier, but serious teams still need compatibility tests, security boundaries, compliance checks, and a rollback plan.