Anthropic grants EU cybersecurity agency ENISA testing access to Mythos 5; agency still lacks access to Mythos 5.1
anthropic
| Source: Techmeme | Original article
Anthropic has granted ENISA testing access to its Mythos 5 model after talks that began in late May, but the agency still lacks access to the newer Mythos 5.1.
Anthropic has granted the EU’s cybersecurity agency ENISA testing access to its flagship large‑language model Mythos 5, the European Commission confirmed after negotiations that began in late May. The move follows months of dialogue between the two parties, but ENISA still does not have access to the newer Mythos 5.1 version, which remains under Anthropic’s control.
The access deal is significant because Mythos 5 has been highlighted for its advanced “hacking” capabilities, raising concerns about how such powerful models might be misused. By allowing ENISA to evaluate the model directly, the EU aims to assess potential security risks, develop mitigation strategies, and shape regulatory guidance for high‑risk AI systems. The commission’s spokesperson for tech sovereignty, Thomas Regnier, described the engagement as “constructive,” underscoring the bloc’s push for greater oversight of frontier AI technologies.
The delayed rollout – more than three months after Anthropic first signalled the possibility of EU access – illustrates the tension between rapid AI innovation and the need for thorough security vetting. It also highlights a broader pattern of European authorities seeking hands‑on insight into models from both Anthropic and rivals such as OpenAI, whose GPT‑6 Astra is reportedly also under ENISA review.
What to watch next: whether Anthropic will extend the same testing privileges to Mythos 5.1, and how ENISA’s findings will feed into upcoming EU AI legislation. Observers will also monitor any further collaboration between EU regulators and AI firms, especially as the bloc tightens its stance on high‑risk models after recent warnings from researchers about existential threats posed by unchecked AI development.
Sources
Back to AIPULSEN