An anonymous AI model that quietly outperformed GPT-5.6 on coding tests has been unmasked as a Z.AI release. The Cl0p ransomware group has now named over 40 victims, including Shell and Philips, in a campaign against PTC software. And Microsoft's August patches fix 421 flaws, including one already under attack.
A cyberattack disrupted Boston Scientific's global order and shipping systems. IBM released open-weight Granite 4.2 models built for enterprise AI agents. And China's Moonshot AI is in talks with Microsoft, Amazon and Google to bring its huge Kimi K3 model to their clouds.
A threat report shows an AI coding assistant driving nearly every step of a live ransomware break-in across eight organisations. OpenAI paused internal work on its Astra model after it could not rule out a "Critical" cyber capability. And Google Cloud says agent security is now the biggest blocker to scaling AI in the enterprise.
Researchers showed that encrypted AI reasoning blocks can leak API keys and passwords across sessions. Google moved its A2A agent protocol under the same open governance as Anthropic's MCP. And Z.ai released GLM-5.3, a 743-billion-parameter open-weight coding model.
A London AI lab's small agent outperformed frontier models from Anthropic and OpenAI at reproducing scientific papers. Starcloud raised $250 million to build data centres in orbit. And a breach at France's tax authority exposed data belonging to 678,000 people.
The Model Context Protocol's stateless core has landed in Amazon Bedrock AgentCore. Microsoft patched a critical, unauthenticated Windows DNS Server flaw that is wormable. And Azure, AWS and Google Cloud all posted accelerating growth, with AWS and Google Cloud launching a joint multicloud collaboration.
