What Altar is
Belgian security company Aikido released Altar, an open-weight model for defensive cybersecurity on customer infrastructure. It is a compressed, customized version of Z.AI’s GLM-5.3 and is used inside Aikido products for customers including Belfius. Aikido reached a reported $1 billion valuation in January.
Why this matters
Security systems inspect unreleased source code, architecture and vulnerabilities. Closed cloud APIs ask defenders to move that material across an organizational boundary. A capable local model changes the calculation by keeping code and findings on premises while automating triage.
Perspectives and European context
Anthropic’s restricted “Mythos” is described as highly capable at finding—and potentially weaponizing—vulnerabilities. Open weights improve auditability and sovereignty but can equip attackers. Aikido’s implied answer is that adversaries already use capable models, so defenders lose more from scarcity. The EU AI Act and the conversion of supercomputers into AI factories make local inference part of a broader push to reduce dependency on U.S. clouds.
Winners, losers and what follows
Banks, defense suppliers and software firms with strict controls gain an alternative. European infrastructure providers gain demand. Frontier-model companies may lose sensitive workloads. Smaller organizations may lose if local deployment demands expensive chips and specialists. Independent testing should compare Altar with closed models on unseen repositories, false-positive rates and rare high-severity flaws. “Runs locally” is a useful property—not proof of safety.
Sources: Reuters on Altar; Reuters on European AI infrastructure.