Anthropic’s Claude Mythos didn’t just raise eyebrows in the AI community. It gave IBM a platform to make its case.
In a commentary published April 9, 2026, IBM argued that the release of Claude Mythos — a model capable of finding and exploiting software vulnerabilities at a level that rivals top human experts — marks a turning point. AI is no longer an experimental tool. It is becoming infrastructure. And once something becomes infrastructure, the rules change.
From Product to Platform to Infrastructure
IBM framed the shift this way: at the product stage, control feels like an advantage. Closed systems move fast, tightly manage the experience, and concentrate value inside a single company. That approach can work early on. Stock Titan
But it stops working when the technology goes deeper.
As Rob Thomas, SVP and CCO at IBM, outlined, software typically graduates from a standalone product to a platform, and then from a platform to foundational infrastructure — and that changes the governing rules entirely. Ibvl
At that point, no single company can anticipate every failure mode, every adversarial use case, or every operational need. Keeping AI development closed becomes harder to defend, not easier.
Why Openness Improves Security
IBM’s argument cuts against a common instinct. Restricting access to powerful systems can look like caution. IBM says it isn’t.
Once AI becomes critical infrastructure, opacity can no longer be the organizing principle for safety. For decades, the most reliable model for secure software has been open foundations combined with serious governance, active maintenance, and broad scrutiny. IBM
IBM argues that open source enables wider scrutiny, which improves security more often than concealment. Stock Titan More researchers, developers, and security defenders examining a system means more weaknesses get caught before they become crises.
This is the same logic that made Linux, the internet’s core protocols, and cloud infrastructure more resilient over time — not less.
Who Participates Shapes What Gets Built
IBM’s argument isn’t purely about security. It’s also about who gets to influence the direction of AI.
Narrow access leads to narrow perspectives. Broad access enables more researchers, startups, governments, and institutions to influence how technology evolves and where it is applied. That builds legitimacy and adaptability. IBM
Arnaud Le Hors, a Senior Technical Staff Member at IBM Research, envisions collaborations between large enterprises on large language models becoming more common IBM — not as a sign of weakness, but as a competitive and structural advantage.
The Enterprise Cost of Closed AI
For companies already deploying AI, IBM’s argument carries a practical edge.
Closed proprietary models create troubleshooting bottlenecks when connected to enterprise vector databases or internal data lakes. When hallucination rates spike or outputs go wrong, teams lack the internal visibility to diagnose whether the error originated in the retrieval layer or the model itself. Ibvl
The spiraling compute costs of continuous API calls to locked models also erode the profit margins these systems are supposed to enhance. Ibvl
IBM’s own approach — built around its open-source Granite model family and the watsonx governance platform — reflects this philosophy. IBM Granite 4.0 became the world’s first open-source language model to achieve ISO/IEC 42001 certification Nemko Group AS, a compliance credential that IBM says removes a major barrier for regulated industries like finance, healthcare, and government.
The Mythos Moment
IBM didn’t miss the significance of Claude Mythos as a catalyst.
Anthropic says Mythos can discover and exploit software vulnerabilities at a level that few human experts can match, and it has launched a gated initiative called Project Glasswing to put those capabilities into defenders’ hands first. IBM
IBM’s read: that’s exactly the kind of capability that makes openness more necessary, not less. The more powerful the model, the more stakeholders need the ability to inspect, question, and improve it.
IBM contends that once AI is infrastructure-scale, openness becomes a design requirement for long-term safety — not a philosophical preference. Stock Titan
The debate over open versus closed AI has run for years. IBM’s position is that Mythos may have just ended it.
