AI news, October 20, 2026: TorchNeuron brings PyTorch to Trainium, OpenCode key breach, Tenstorrent raises $1.5B
Five stories from October 17โ19, 2026 (US time) for people who build software with AI tools or pay for AI APIs. Each item links to the original announcement or reporting; numbers are quoted from those sources, not estimated here.
Amazon's TorchNeuron lets PyTorch code run on Trainium as a standard device
AWS announced TorchNeuron, a native PyTorch backend for its Trainium2 and Trainium3 chips. In the PyTorch blog post, moving a model to Trainium is shown as .to('neuron'), with eager mode, torch.compile, the standard distributed APIs and torch.profiler supported. AWS says it goes live as a public beta on October 20 and will be open-sourced under Apache 2.0 at aws-neuron/torch-neuronx. Reuters reports more than 30 early customers are already using it.
Why it matters for makers: if you train or serve open models, Trainium becomes easier to try without rewriting PyTorch code.
Sources: PyTorch blog, Reuters via Investing.com
OpenCode says a breach exposed provider API keys for 1.4% of Console users
The open-source coding agent OpenCode said a large-scale distillation attack hit 1.4% of its Console users, all of whom had linked external providers such as Anthropic, OpenAI or Google. BleepingComputer reports that the notice to affected users puts the earliest possible key exposure at October 17, 00:30 UTC, and that some users reported unauthorized charges of up to $1,700 and suspended Google Cloud projects. OpenCode said it emailed every affected user and is working with the three providers.
Why it matters for makers: any tool that stores your AI provider keys is part of your attack surface. Set spending limits and rotate keys you have pasted into third-party dashboards.
Sources: BleepingComputer, TechCrunch
Tenstorrent closes a $1.5B Series E at a $6.2B valuation
AI chip company Tenstorrent, led by CEO Jim Keller, closed $1.5 billion in Series E funding at a $6.2 billion post-money valuation. The round was led by Trace Capital and Innovation X, with investors including ARK Invest, Fidelity and Morgan Stanley. The company says the money will go to deploying its Galaxy servers at scale for open-weight models and to the RISC-V CPU and AI core IP it licenses to chip designers.
Sources: Tenstorrent newsroom, TechCrunch
Anthropic commits 5% of its shares to a new public benefit fund
Anthropic said that, subject to stockholder approval, 5% of its shares will go to the Anthropic Public Benefit Fund, with another 2% possible at the board's discretion for every 50% rise in company value above a set baseline. The fund is a wholly owned LLC rather than a nonprofit, and Anthropic says none of its equity will be used for lobbying or political giving. Together with earlier founder and employee pledges, Anthropic says roughly 20% of its shares are now dedicated to the public good; it names AI's effects on the labor market as one possible focus.
Replit makes OpenAI the default for Replit Agent, Claude stays selectable
Replit CEO Amjad Masad said at TechCrunch Disrupt 2026 that the change was "a decision based on both cost, partnership, and performance". He later clarified on X that Replit remains model agnostic: OpenAI is now the preferred partner powering Agent by default, and users can still choose Anthropic and other supported models. Replit's docs say manual model choice is available in Power and Max modes, while Free Mode uses automatic routing.
Sources: TechCrunch, Replit docs