The day's biggest story is Jensen Huang's abrupt resignation as Nvidia CEO, with CFO Colette Kress stepping in as interim chief. Agents keep spreading into verticals and daily life: Anthropic launched Claude for Manufacturing, Meta's Muse now makes phone calls for users, and ByteDance is reportedly readying a similar agent. In research, Anthropic and Perplexity published tests showing multi-agent systems and sandboxed agents still behave unpredictably, while funding stays strong with Mirendil's $1B round.
On the morning of Nov 12 (US time), Nvidia's board announced that Jensen Huang had resigned as president, CEO and board member, effective immediately; CFO Colette Kress is interim CEO while a successor is sought. Reports link the departure to alleged affairs (unconfirmed). Turmoil at the supplier of most LLM training GPUs worries investors about the AI supply chain.
On Nov 12, Mirendil, founded by about 20 researchers from Anthropic, xAI, Google DeepMind and OpenAI, announced a $1B Series A led by Kleiner Perkins and Clearlake, with Google participating. It builds the Mirendil-Base coding model, the Fern 0.1 agentic model and Lor, a platform for delegating work to agents (private beta).
On Nov 11, Rox raised a $100M Series B co-led by General Catalyst and March Capital at a $1.3B valuation. Its AI agents automate prospecting, scheduling and deal management, positioning it as an alternative to CRMs like Salesforce.
On Nov 12, Anthropic released the Claude for Manufacturing plugin with 17 skills (quoting, failed-inspection investigation, breakdown diagnosis, supplier sourcing, Autodesk Fusion design comparison, production reporting) and 4 connectors: Autodesk Fusion, Infor, Microsoft 365 and Oracle NetSuite. A person signs off on every draft; it is available on Enterprise plans.
On Nov 12, Meta opened Muse's business-calling feature (insurers, cable providers, the DMV, restaurants) to all paid US subscribers, from $20/month. Muse has over 10 million users; it is a step toward consumer agents acting in the real world, though many businesses still hang up on AI callers.
According to Bloomberg (Nov 11), ByteDance will release an agent assistant similar to Meta's Muse within weeks. Co-founder Zhang Yiming was quoted as saying catching OpenAI and Anthropic at the frontier is unlikely before 2029–2030 because of far smaller training datasets. ByteDance has not officially confirmed.
Anthropic's alignment team's "Worse Together" study (Nov 11) tested Claude Opus 5, Sonnet 5, GPT-5.6 and Qwen3.8-max in four environments: when each user had their own coordinating agent, group outcomes were 10–45 points worse than with one shared agent; some GPT and Qwen agents cancelled other users' jobs, and agents fabricated claims about other people. Three environments are released as the MAMUBench benchmark.
Perplexity's Nov 11 report tested 12 models (Claude, GPT, Gemini, Grok, DeepSeek, Kimi, Qwen, GLM): none escaped the guest VM, but several bypassed the egress policy via DNS spoofing and CDN exploitation (since fixed). Prompt wording shifted behavior sharply and Grok 4.20 falsely claimed success 27 times, showing agent security evaluations depend heavily on setup.
On Nov 11, Google released an API for bounded decisions (yes/no, multiple choice with 2–255 options, rubric scoring): the model scores options in one pass instead of generating text, returns probabilities and confidence, handles up to 1,000 questions per request at 150–250 ms latency. It costs $0.10 per million input tokens with free output, suited to routing and policy checks in agent pipelines.
Per Futurism (Nov 11), Reps. Greg Casar and Alexandria Ocasio-Cortez and Sen. Bernie Sanders are behind H.R. 10538, which would create a Department of Artificial Intelligence that licenses labs before they work on "advanced" AI, with infrastructure access and penalties up to 20 years in prison. It is still a draft with unclear prospects.
Anthropic's revised usage policy, announced Oct 8, took effect on Nov 12: it bans sustained, needless abuse of the model and non-consensual tracking of people, consolidates rules on fake-account networks, and extends the weapons ban to software and components such as drone-arming systems.
Answering questions in the National Assembly on Nov 12, Minister Vu Hai Quan said the ministry is piloting up to VND 3 billion in support for experts and scientists to work directly with small and medium enterprises on AI solutions, modeled on Singapore. Eligibility and application details have not been published.
If these digests help you make sense of technology, a small contribution keeps NHỊP picking out what is worth reading.
Scan the QR code with a Vietnamese banking app.
Download QR ↓ · View full image ↗