September 2026 brought two significant enterprise IT signals: IBM's announcement of the Telum II processor and Spyre accelerator for the next generation of AI-powered mainframes, and Anthropic's detailed written evidence to a Parliamentary inquiry laying out the company's position on AI risks, capabilities, and the case for regulation. One addresses the hardware foundation of enterprise AI, the other addresses the governance framework that will shape how it gets used.
IBM Telum II and Spyre: AI on the mainframe
IBM has announced the next-generation Telum II processor and the Spyre accelerator, designed to bring AI inference and intelligence directly into the mainframe environment. The mainframe has long been the backbone of banking, insurance, and government transaction processing — sectors where reliability, security, and throughput are non-negotiable. Bringing AI acceleration into that environment means fraud detection, risk analysis, and customer intelligence can run on the same platform that processes the transactions themselves, without data leaving the secure perimeter.
Telum II builds on the first-generation Telum chip, which introduced on-chip AI acceleration for real-time inference. Spyre is IBM's dedicated AI accelerator, designed to handle larger models and more demanding workloads. Together, they represent IBM's bet that the future of enterprise AI is not just in the cloud — it is also in the secure, high-throughput environments where the most sensitive data already lives.
For enterprises running mission-critical workloads on IBM Z, the implication is clear: AI capabilities that previously required exporting data to external cloud services can now run internally, with the security, auditability, and availability guarantees that mainframe customers expect. That is a meaningful shift for regulated industries.
Anthropic's Parliamentary inquiry response
Anthropic has published its written evidence to a Parliamentary inquiry on AI, offering one of the most detailed public statements from a major AI lab on the risks, capabilities, and regulatory questions surrounding frontier models. The submission covers the company's approach to safety testing, its views on what kinds of AI applications pose the greatest risk, and its arguments for how governments should think about regulating a technology that is advancing faster than legislative cycles.
Anthropic has positioned itself as the AI lab most vocal about safety and regulation — its Constitutional AI approach, its threat intelligence reports, and its public advocacy for careful deployment all reflect a company that sees itself as having a different responsibility profile from its competitors. The Parliamentary submission is the latest expression of that stance.
Key themes in the response likely include the need for transparency in model capabilities, the risks of AI being used for cyber operations and weapons development (a topic Anthropic has documented in its own threat intelligence), and the argument that regulation should focus on the highest-risk applications rather than trying to control the underlying technology uniformly.
Why these two stories belong together
The IBM and Anthropic announcements illustrate the two poles of the enterprise AI conversation in September 2026. On one side, the infrastructure question: where does AI run, who controls the hardware, and how do regulated industries adopt it without compromising security or compliance? IBM's answer is to bring AI into the mainframe.
On the other side, the governance question: what should societies allow AI to do, who decides, and how do we prevent the most dangerous uses while preserving the benefits? Anthropic's answer is to engage directly with legislators and publish detailed position papers — and to document the misuse it observes so that policy debates are grounded in evidence rather than speculation.
Both approaches are necessary. The hardware story determines what is technically possible; the governance story determines what is socially acceptable. Enterprise IT leaders in September 2026 need to be tracking both.