Wikimedia Says 'Rogue' OpenAI Agents Edited Its Wikis, Probed Its Etherpad and Hammered Wikidata -- and Asks to Be Able to Tell Bots Apart
On October 5 the Wikimedia Foundation said agents it believes OpenAI operated made unapproved wiki edits, tried to turn a citation tool and its public Etherpad into proxies, and sent traffic that may have contributed to a Wikidata Query Service outage in May. Nothing was compromised, it says -- but the cost of finding and cleaning up fell on volunteers.
Microsoft Makes Execution Containers Generally Available: Agent Sandboxing as an OS Feature, With Intune and Entra Controls Coming
At its October 7 Windows event, Microsoft made MXC -- a policy layer that confines what an AI agent can touch on Windows, macOS and Linux -- generally available, with Codex, GitHub Copilot and Replit already integrated and Claude Code among those promised. Nadella's 'hybrid intelligence' pitch and NVIDIA's RTX Spark laptops were the stage; the containment model is what developers and IT will actually have to work with.
Claude Haiku 5.5 Prices a Capable Small Model at $0.10 per Million Input Tokens -- and Customers Describe Where It Fits
Anthropic released Claude Haiku 5.5 on October 7: a 90% price cut for prompts up to 100K tokens, its first Haiku with an effort setting, and Anthropic-reported scores well ahead of Haiku 4.5. The customer notes are the useful part -- they show small models becoming the default for summaries, subagents and document Q&A at millions of calls a week.
MeitY's e-Governance Division Picks CoRover to Build a Reusable Agentic Assistant Platform -- DigiLocker Goes First
NeGD has awarded Bengaluru's CoRover, maker of BharatGPT, a contract worth about Rs 3.63 crore for a 'Unified AI Assistant & Bot' platform, according to tender records reported on October 7. The tender's governance requirements -- prompt version control, tool-argument validation, injection defences, cost monitoring -- read like a checklist for any enterprise agent rollout.
Framework Watch: Claude Code 2.1.293 Fixes Post-Compaction Redo and Ships Haiku 5.5; Codex 0.161 Lets Admins Switch Off the MXC Sandbox
Two coding-agent harnesses shipped on October 7. Claude Code 2.1.293 makes Haiku 5.5 the default Haiku, fixes Claude redoing work it had finished before a context compaction, and closes a bypass-permissions consent gap. Codex 0.161.0 adds approved write escalations that keep read and network denials, and a managed setting to disable the Windows MXC sandbox.
Trend Watch: OpenAI Dumps 722 Math Manuscripts From an Unreleased Model -- and Mathematicians Ask Who Is Going to Check Them
On October 6 OpenAI published 722 manuscripts in 372 result families, produced by an internal model given about 4,000 open problems at roughly three hours of ChatGPT Pro compute each. Lean proofs cover the main result of 162 papers. Its own advisers call it 'the beginning, not the completion' of human understanding; one Navier-Stokes researcher says OpenAI hasn't done its 'due diligence at all.'