Belgian cybersecurity company Aikido Security has launched an open-weight AI model designed to run inside customers’ own infrastructure, reflecting growing demand for security tools that can analyze sensitive software without sending proprietary code to external AI providers.
The Ghent-based company introduced Altar, its first open-weight security model, as technology companies increasingly weigh the capabilities of frontier AI against concerns about where source code and security data are processed.
Aikido says Altar is built to provide frontier-grade defensive security inside infrastructure controlled by the customer. The model powers its Aikido Machine autonomous penetration-testing product, including deployments in fully air-gapped environments.
Sensitive code no longer has to leave the network
The architecture addresses a specific problem facing security teams adopting generative AI.
Using externally hosted models for vulnerability analysis can require sending source code, configuration data or information about an organization’s attack surface to a third-party inference service. Aikido says Altar allows its security system to operate without sending an organization’s sensitive context to an external AI provider.
Reuters reported that Aikido’s model is intended to be used locally for cybersecurity applications, as security providers respond to demand for tools that can operate without exposing confidential code to cloud-hosted models.
That makes the product particularly relevant for organizations operating under strict confidentiality, sovereignty or infrastructure-isolation requirements.
A European unicorn is moving deeper into autonomous security
Aikido was founded in 2022 by a team including CEO and CTO Willem Delbare, COO Roeland Delrue, and CMO Felix Garriau.
The company raised $60 million at a $1 billion valuation in January 2026, in a round led by DST Global. At the time, Aikido said revenue had increased fivefold over the previous year, with roughly half coming from the United States.
Aikido now says it has raised $85 million and protects more than 150,000 teams, with products covering code, cloud infrastructure, runtime security and autonomous penetration testing.
The company’s positioning has also shifted beyond conventional vulnerability scanning. Its stated goal is “self-securing software” in which systems continuously test and improve their own defenses.
Local models could become part of the security stack
Altar points toward a larger infrastructure question for enterprise AI.
For many software teams, cloud-hosted models offer greater convenience and access to rapidly improving capabilities. Security workloads are different because the material being analyzed may reveal exactly how an application can be attacked.
That gives local inference a practical advantage even when it requires more infrastructure to operate.
For European security companies, sovereignty is also becoming a product feature rather than only a regulatory slogan. If customers increasingly demand AI systems that can work inside private environments, security vendors may compete not only on detection accuracy but on where computation happens and who can access the data surrounding it.
Aikido’s bet is that the next generation of cybersecurity AI will not always live behind somebody else’s API.
For some of the most sensitive enterprise workloads, the competitive advantage may be keeping the model—and the code it is examining—inside the building.
