Google Launches Autonomous Enterprise Agents at Gemini at Work 2026
Google Cloud CEO Thomas Kurian took the stage at the company's Gemini at Work 2026 event on October 8 to unveil what may be the most ambitious enterprise AI product yet: a unified Gemini agent that works across Gmail, Drive, Docs, Sheets, Slides, Chat, and Calendar — and gets its own dedicated Workspace account to do it.
Each agent receives a unique identity with its own Gmail address, calendar, and Drive storage. Access is tightly controlled: the agent can only see information that team members explicitly share with it. Google says the platform uses cryptographic agent identities to verify that an agent is who it claims to be, a first-of-its-kind approach to enterprise AI authentication.
These aren't quick one-shot assistants. Google's agents are designed to persist across hours or even days of work, orchestrating sub-agents and reaching across organizational boundaries. The system also integrates with Microsoft 365 and Slack, and can route requests to Anthropic's Claude models alongside Google's own. Maximum runtime is set at seven days.
Industry-specific agents for financial services and legal are already in preview, with healthcare, retail, and government versions on the way. The platform is currently in private preview, with broader availability planned for Workspace Business and Enterprise customers.
Sources: Google Cloud Blog, 9to5Google
OpenAI's 722 AI-Generated Math Papers Ignite Fierce Academic Backlash
On October 6, OpenAI published a repository containing 722 manuscripts grouped into 372 families of mathematical results, all generated by an unreleased frontier model. The release included a companion library in the Lean proof assistant. It was the largest single dump of AI-generated mathematical work in history.
The reaction from the mathematics community was swift and sharp. The newly formed Association for Human Mathematics (AHM) issued a blistering statement, reposted on Fields Medal winner Terence Tao's blog, calling the release "not a demonstration of scholarship, but a demonstration of power." The group urged mathematicians to discontinue their work with OpenAI and argued that the company showed "total disregard for the norms of scientific research."
Not everyone agrees. Some researchers expressed excitement about the new results opening vast areas for exploration. Mathematician Daniel Litt pushed back on the backlash, writing: "If we want to know the answers to these math questions, I see no reason why we should ask the company to keep them secret from us."
The controversy highlights a growing tension between AI companies eager to demonstrate frontier capabilities and the academic communities whose norms they bypass. OpenAI tested its model against roughly 4,000 mathematical problems, but the release appears to have circumvented guidance from its own outside advisory group.
Sources: Terence Tao's Blog, Fortune, Scientific American
Anthropic Launches Cyber Mission to Defend Critical Infrastructure
Anthropic announced its Cyber Mission on October 8, a sweeping initiative to apply AI to the defense of critical infrastructure and open-source software. The program launches with 11 founding partners, including CrowdStrike, Palo Alto Networks, Dragos, Deloitte, Accenture, Booz Allen, PwC, and Rockwell Automation.
The centerpiece is the Critical Infrastructure Defense Program (CIDP), which gives participating organizations access to Claude models, on-site engineers, and threat research. The focus is on operational technology — the control systems behind power grids, water systems, and transportation networks — which often can't be taken offline for patching and may carry unresolved vulnerabilities for years.
Anthropic also launched OSS Scanner, a free tool that provides regular AI-powered security scans for open-source projects. The company claims a true-positive rate above 90%, though it acknowledges that some reports may contain inaccuracies such as incorrect severity ratings. These are model-generated with no human review.
The initiative builds on the company's earlier Project Glasswing, which has been folded into a broader Cyber Verification Program. Anthropic frames the effort as a response to state-sponsored adversaries who have "established footholds across multiple technology sectors."
Sources: CryptoBriefing, NewsBytes
Mistral Debuts "Le Chonk": Europe's First Trillion-Parameter AI Model
French AI company Mistral AI opened a public preview of Mistral Large 4 on October 6, internally nicknamed "Le Chonk." With 1.05 trillion total parameters and roughly 49 billion active per token, it's the first European-origin model to reach the trillion-parameter frontier.
The model is natively multimodal, with a 1.6 billion parameter vision encoder that processes images, diagrams, and documents. Mistral says it surpasses closed frontier models on visual grounding tasks. It covers more than 160 languages, including every official EU language.
Training took approximately two months on 4,000 Nvidia Grace Blackwell GPUs in Mistral's own European data centers — a notable fact for customers with data sovereignty requirements. The API is live now, with open weights promised by late October pending additional security testing. One report suggests the target date is October 27.
The release marks a significant milestone for European AI sovereignty. While the licensing terms haven't been announced, the promise of open weights at this scale could reshape the competitive landscape for enterprise and research users alike.
Sources: VentureBeat, BetaNews
In Brief
UN calls for urgent AI guardrails: UN High Commissioner for Human Rights Volker Türk warned that "the clock on AI regulation is ticking," calling for mandatory human rights safeguards that adapt as the technology evolves. (UN News)
India plans AI safety rules: IT Minister Ashwini Vaishnaw announced on October 8 that India will publish a consultation paper on AI safety and deepfake regulations within a month. (CareerIndia)
Claude Haiku 5.5 ships: Anthropic released Claude Haiku 5.5, the newest member of its model family. It features adjustable effort settings and a 1M context window, designed as a fast, cost-effective agent helper alongside Opus 5.5 and Sonnet 5.5. (Evertune AI Model Tracker)