Aikido, a Belgian cybersecurity company, has launched an open-weight artificial intelligence model aimed at security teams that want to run their AI in-house rather than in the cloud. The model, called Altar, was announced on Sunday and quickly picked up by Reuters, which reported the launch as part of growing demand for local security tools.
For the latest breaking AI news and analysis, visit AI Buzz Wire.
Built for Sovereign Security
The company is positioning Altar around what it calls sovereign security: an AI model whose weights are openly available and which can be deployed on-premise, inside an organization's own infrastructure. That matters for security operations centers, financial institutions, and government agencies where sending logs, code, and incident data to a third-party cloud model is a non-starter — either because of regulation, confidentiality, or the simple reality that the data being analyzed is exactly the data an attacker wants.
Reuters framed the launch as a response to rising demand for local tools, and the on-premise angle puts Aikido in a small but growing group of security vendors building products around open-weight models rather than API-only services.
A Vendor That Benchmarked Its Way Here
Altar did not come out of nowhere. Aikido has spent the past months publishing research on how well commercial AI models actually perform at security tasks. In July, the company benchmarked 13 AI models on rediscovering known CVEs — testing whether models could find real, documented vulnerabilities. In August, it published results from an evaluation it says consumed 11.7 billion tokens, concluding that models from Z.ai (GLM-5.3) and DeepSeek had reached frontier-level performance on cybersecurity work.
That research matters for context: Aikido has been arguing publicly that open-weight models from Chinese labs are competitive at security tasks, and Altar is the logical product response — build the security product on a model the customer can host, audit, and control.
Why On-Premise AI Security Is Having a Moment
The launch lands amid a broader shift in how enterprises consume AI. Security teams face a specific version of the dilemma: the same confidentiality constraints that make cloud AI awkward are most acute for the people whose job is handling breaches. An open-weight model that runs inside the perimeter lets a security team use AI for triage, vulnerability analysis, and incident response without exfiltrating sensitive data — and without the model's behavior changing under them after a provider-side update.
It also reflects a European angle. With EU data-protection rules and increasing scrutiny of where data travels, demand for AI that never leaves the building has become a selling point in its own right, particularly for regulated industries.
What Is Not Yet Known
Initial coverage from Reuters and other outlets focused on the strategic launch rather than technical specifics. Details such as Altar's parameter count, which base model it derives from, its license terms for commercial use, and independent benchmark results were not part of the first-day reporting. As with any vendor-published launch, security buyers will want to see the evaluation methodology behind any performance claims — a standard Aikido itself applied aggressively to other vendors' models in its own published benchmarks.
The launch is nonetheless notable as a sign of where the AI market is heading: away from a pure API economy and toward organizations choosing, hosting, and owning the models their critical workflows depend on.
Stay Ahead of AI
For continuous coverage of the AI industry's most important developments, bookmark AI Buzz Wire and never miss a breaking story.
Read more AI news
