Lakera Red and Lakera Guard AI Security

Lakera is an AI security company founded by AI researchers with backgrounds at Google and Meta, operating from Zurich and San Francisco. Check Point Software Technologies announced its acquisition in September 2025, and Lakera now forms the basis of Check Point's Global Center of Excellence for AI Security while keeping its own brand and site. It addresses prompt attacks, data leakage and unsafe tool use in enterprise GenAI applications, agents and MCP-connected systems.

The platform covers the AI lifecycle. Lakera Red runs risk-based red teaming before and after deployment, simulating direct and indirect prompt injection, data exfiltration and harmful outputs, and scores results with a normalized risk score and attack success rate. Lakera Guard, now offered as AI Agent Security, discovers agents and MCP tools, assesses their risk and enforces runtime guardrails through an API-first, model-agnostic service. A Workforce AI Security module governs employee use of AI apps. Its threat intelligence draws on the Gandalf game and a community of over 1 million hackers. Named users include Dropbox and Pearson.

• Risk-based red teaming: Lakera Red simulates direct and indirect prompt injection, data exfiltration and harmful suggestions against apps and agents.
• Risk scoring: reports a normalized 0-1 risk score and attack success rate per scenario for comparison over time.
• Agent discovery: finds agents and MCP-connected tools, including unregistered ones, and assesses their configuration and data access.
• Runtime guardrails: blocks prompt attacks, data leakage and unsafe tool use with sub-50 ms latency.
• Workforce AI Security: discovers unsanctioned AI use in apps and browsers and applies policies by user, app and action.
• Expert red teaming service: adversarial testing delivered by Lakera specialists.

Vietnamese banks and e-wallets can red team customer-facing chatbots and agents before launch and keep runtime guardrails against prompt injection. Enterprises already using Check Point firewalls can extend policies to employee use of public AI tools. Telecom and e-commerce firms can protect multilingual support assistants from data leakage.

Security teams test GenAI apps and agents against realistic attacks before launch and track the risk score over time. Runtime protection adds under 50 ms of latency according to Lakera, with a stated production false positive rate of 0.01%. Detection covers over 100 languages, useful for multilingual customer service. Central policies apply without changing models or prompts.

Enterprises are moving from chatbots to autonomous agents connected to internal tools through MCP, which widens the attack surface for prompt injection and data exfiltration. Large security vendors are acquiring AI security specialists, and buyers in Southeast Asia increasingly expect AI testing and runtime controls to come from their existing security suppliers.

Check Point announced the acquisition of Lakera on 16 September 2025, with closing expected in Q4 2025; the site now carries Check Point copyright and legal pages. The homepage lists Lakera Inc at 282 2nd Street, San Francisco. Lakera runs the Gandalf adversarial game and cites OWASP, WEF and Snyk as recognition or partners.