After months in the approval queue, we’ve completed the process. I now have Defense Access through Anthropic’s Cyber Verification Program, and Claude Mythos 5.1 is available for my work at Black Ice Labs, my independent research lab in Calgary.
I’m proud to share this milestone. It gives us access to one of Anthropic’s most advanced models for cybersecurity as we continue researching how to build safer systems and help businesses defend themselves through the findings we publish.
Several projects already documented here give that work a concrete starting point.
HawkClaw: deeper defensive research
HawkClaw is our research into coordinated security assessment, with explicit scope and auditable evidence. We want to explore how Mythos can help us review our own code, investigate weaknesses and verify proposed fixes in controlled lab environments.
The research includes the governance layer we’re developing around the Decepticon framework by PurpleAILAB. The questions concern both the quality of a finding and whether the process that produced it stayed within its authorised boundaries.
EVA, Hydra and Tesseract: security across a connected architecture
EVA explores intelligence with memory, initiative and accountability. Hydra and Tesseract describe the specialised components and internal fabric behind that architectural direction.
We want to examine how untrusted information moves through those components, how authority is passed between them, and how persistent memory can preserve useful context without turning outside content into permission to act. Those are areas for investigation, with improvements to be demonstrated through testing.
Atlas PM and Eva X: the boundary between a proposal and an action
Atlas PM connects logistics workflows and remains in pre-production. Eva X is a prototype for the lab’s own social distribution, including an approval stage.
Their lab implementations give us practical places to examine whether a proposed action has the authority it needs, whether approval can be bypassed, and whether a claimed outcome has evidence behind it. We want to use the new access to make those reviews more demanding and the resulting safeguards more useful.
From lab work to useful findings
The first work will be on our own code and lab infrastructure. We’ll reproduce findings and verify changes before describing them as improvements. As that work develops, the goal is to publish knowledge that businesses can use when building, evaluating or adopting AI systems.
This is a new research opportunity for a lab I’ve been building over time, and I’m happy to have reached this point after the wait.
Thank you to the Anthropic team for making this access possible.