The European Commission has begun assessing the potential risks posed by Anthropic’s controversial AI model Mythos, signaling that European regulators are increasingly alarmed about how advanced artificial intelligence systems could destabilize financial networks and accelerate cyberattacks against critical infrastructure.

European Economic Commissioner Valdis Dombrovskis said Monday that European Union officials had already held discussions with Anthropic and received technical briefings about the model’s cyber capabilities as authorities weigh whether existing EU regulations are sufficient to address the emerging risks.

“The commission representatives met with Anthropic and was briefed on technical details around cyber capabilities and the risk of this Mythos preview, so we are currently assessing possible implications in light of the EU policies and legislation,” Dombrovskis told reporters.

The comments mark one of the clearest indications yet that European policymakers are moving beyond theoretical debates about artificial intelligence and beginning to confront what regulators increasingly view as immediate operational and national security risks tied to advanced AI systems.

Anthropic’s Mythos model, which cybersecurity experts say is capable of identifying vulnerabilities in software code at unprecedented speed and scale, is at the center of those concerns. Security researchers fear such systems could dramatically compress the timeline between the discovery of software flaws and their exploitation by malicious actors.

Traditionally, hackers often required weeks or months to weaponize newly discovered vulnerabilities. Advanced AI systems, however, are increasingly believed capable of automating large parts of that process, potentially enabling sophisticated attacks within hours of a flaw becoming public.

That shift is raising alarm across governments, banks, and critical infrastructure operators globally. European regulators appear especially focused on the financial sector, where the growing integration of AI into banking operations is expanding the potential attack surface for cybercriminals and hostile state actors.

Although Mythos has not yet been made available to European banks, officials appear concerned that the technology’s capabilities could still indirectly influence the threat environment by empowering hackers, ransomware groups, or state-backed cyber operations targeting financial institutions.

The scrutiny also comes as the European Union aggressively expands its oversight of artificial intelligence under its landmark AI Act, which introduces some of the world’s strictest rules governing high-risk AI applications.

Under the framework, systems deemed capable of threatening public safety, financial stability, or critical infrastructure could face heightened transparency, compliance, and risk-management obligations.

European authorities are increasingly trying to determine whether frontier AI models with offensive cyber capabilities should fall into those categories.

The discussions with Anthropic are also seen as part of a global shift in regulatory thinking. For much of the past two years, policymakers focused primarily on issues such as misinformation, copyright disputes, and labor disruption linked to generative AI. More recently, however, attention has rapidly pivoted toward cybersecurity and systemic infrastructure risk.

Banks in particular have become a central point of concern. Financial institutions operate massive interconnected digital ecosystems containing sensitive consumer data, payment systems, and real-time transaction infrastructure. A highly capable AI-assisted cyberattack targeting those systems could potentially trigger cascading disruptions far beyond a single institution.

Regulators in several jurisdictions have already begun quietly reassessing cyber response frameworks as AI models become more advanced. In the United States, cybersecurity officials are reportedly considering dramatically shortening deadlines for patching critical software vulnerabilities amid fears that AI-powered hacking systems could sharply accelerate exploitation timelines.

Meanwhile, financial institutions globally are racing to test their own defensive AI systems while simultaneously restricting employee access to certain external models over data security concerns.

The growing unease surrounding Mythos illustrates a broader paradox emerging in the AI race. The same technologies being marketed as productivity breakthroughs for coding, automation, and enterprise software development are also creating entirely new categories of cyber risk.

That tension is becoming especially pronounced in Europe, where regulators have historically taken a more cautious approach toward large technology platforms and data governance than their American counterparts. The European Commission’s engagement with Anthropic suggests Brussels does not intend to remain passive as frontier AI systems evolve into increasingly powerful cyber tools.

The scrutiny appears to foreshadow tougher oversight requirements around model deployment, capability disclosures, and access restrictions, particularly for systems capable of offensive cybersecurity applications. It is also seen as another sign that the AI boom is rapidly becoming inseparable from the next phase of the global cybersecurity arms race.

Amazon is formally expanding access to Anthropic’s Claude Code and OpenAI’s Codex across its corporate workforce, marking one of the clearest signs yet that even the world’s largest cloud companies are increasingly relying on outside artificial intelligence systems to accelerate software development.

The rollout, announced internally by Amazon Vice President of Software Builder Experience Jim Haughwout, signals a major shift in the company’s AI strategy. Rather than relying primarily on its internally developed Kiro coding platform, Amazon is now embracing a broader ecosystem of AI coding assistants as competition intensifies across the technology industry.

The move also underpins how generative AI has rapidly evolved from an experimental tool into a foundational layer of software engineering infrastructure inside large corporations.

“To help you invent more for customers, we are expanding the agentic AI tools available to you,” Haughwout wrote in a memo to staff obtained by Business Insider.

According to the note, Claude Code is being made available immediately to Amazon employees, while OpenAI’s Codex will begin rolling out company-wide on May 12. Both systems will operate through Amazon Bedrock and Amazon Web Services infrastructure, allowing the company to maintain centralized control over computing resources, data handling and security compliance.

“Both run on Bedrock, where all inference runs,” Haughwout wrote. “Both will have easy install for all Amazon builders.”

The decision reflects a growing reality inside Silicon Valley: companies can no longer afford to restrict engineers to internally developed AI systems if rival tools are proving more effective in boosting productivity.

For months, Amazon engineers had reportedly pushed for wider access to Claude Code, which many developers viewed as more capable for certain programming tasks than Amazon’s own Kiro platform. Until recently, use of Claude Code for production work reportedly required special approvals, creating frustration among teams trying to accelerate development cycles.

The broader rollout indicates that Amazon leadership ultimately concluded that limiting access to best-performing AI tools posed a greater risk than opening the ecosystem.

An Amazon spokesperson confirmed the company is now “standardizing” access to Claude Code and Codex across its workforce.

“At Amazon, we’ve long held there’s no one-size-fits-all approach to how our teams innovate,” the spokesperson said. “Our builders are using Kiro for agentic coding, and now with both Claude Code and Codex running on AWS, we are making additional tools available as well.”

The spokesperson added that Kiro remains heavily used internally, saying roughly 83% of Amazon engineers still “primarily” use the company’s in-house platform.

Still, the significance of the announcement extends far beyond developer preferences. It highlights how the AI arms race is reshaping alliances across the technology sector, where competitors increasingly depend on one another for infrastructure, models, and compute power.

Amazon has spent aggressively to deepen relationships with both Anthropic and OpenAI in recent months. In February, the company announced a major partnership with OpenAI involving investments of up to $50 billion. In return, OpenAI agreed to expand use of Amazon’s Trainium AI chips and collaborate with AWS on customized AI services and infrastructure.

Also, Amazon has significantly expanded its backing of Anthropic. In April, the company pledged up to an additional $25 billion investment in the startup, on top of the $8 billion it had already committed. Anthropic also agreed to purchase $100 billion worth of Amazon Trainium chips over time.

Those arrangements are widely viewed as part of Amazon’s broader effort to challenge Nvidia’s dominance in AI computing while cementing AWS as the backbone of enterprise AI deployment. The rollout of Claude Code and Codex through Bedrock serves that strategy directly. Even when employees use outside models, Amazon still keeps workloads, inference activity, and cloud consumption within its own infrastructure environment.

That distinction is increasingly important as cloud providers battle for long-term control of what many analysts view as the next major computing platform.

The shift also exposes how AI coding assistants are transforming software engineering itself. Modern AI systems are no longer limited to generating snippets of code. They can now debug applications, write documentation, conduct testing, suggest architectural changes, and autonomously complete increasingly complex engineering workflows.

That evolution has triggered growing anxiety across the technology workforce. Several AI executives and investors have warned that coding automation could eventually reduce demand for entry-level programmers and junior developers. Amazon, however, appears to be positioning the technology more as a force multiplier than a replacement mechanism.

The company continues to hire engineers aggressively while integrating AI deeper into development workflows. AWS CEO Matt Garman recently said Amazon was hiring “just as many software developers as we ever had,” while arguing that future engineers would need broader systems thinking and problem-solving capabilities rather than narrow coding expertise.

The internal rollout also comes as investors increasingly scrutinize whether the hundreds of billions of dollars being poured into AI infrastructure are producing measurable productivity gains.

Amazon’s integrating Claude Code and Codex across its engineering organization is expected to help accelerate software deployment, reduce development bottlenecks, and improve operational efficiency at a time when hyperscalers are racing to justify unprecedented capital expenditures on AI infrastructure.

More broadly, the move signals that the future AI landscape may not be dominated by single-model ecosystems. Instead, major technology firms appear to be moving toward multi-model environments where companies combine proprietary systems with specialized external tools, while competing fiercely for control of the cloud infrastructure layer underneath.

In that contest, Amazon is making clear that it is less concerned about whose model engineers use than about ensuring the entire AI economy ultimately runs on AWS.