Mate Security launches Gamebooks to govern AI agents in security investigations

A Gamebook describes investigative intent rather than an execution path. It sets out what must be investigated, what evidence must be established, which conditions should redirect the investigation, which actions an agent may take, and when it has to escalate, stop or request approval. How the agent reaches that outcome is left open, according to the company.

The architecture separates intent from execution. An orchestrator reads an incoming investigation and assembles the relevant Gamebooks. Capabilities give agents reusable security skills not tied to any vendor’s product. Flows handle actual contact with tools and systems, so agents never hold open access to real systems. Grounding comes from Mate’s Security Context Graph, an earlier product that holds an organization’s current state and its history of prior decisions.

Mate said the same Gamebook keeps running through changes such as replacing a security tool or absorbing an acquired company’s stack, because only the execution layer has to adapt. Analyst turnover is handled the same way: decisions made by a departed analyst remain in the context graph with their reasoning. Customers can write their own Gamebooks in natural language, convert existing playbooks into investigative intent, or extend those Mate ships. Proprietary tools and data can be plugged in, while Mate keeps the underlying agent engineering, evaluation and testing.

Every investigation also feeds the loop. Evidence, relationships and outcomes are stored back in the context graph. Useful patterns can be promoted into new detections, and noisy detections can be tuned against what investigations actually turned up. Oren Saban, co-founder and chief product officer, said the move to agentic investigations “requires a different architecture,” one that lets AI reason and adapt while staying anchored to how a particular organization investigates. Security teams should not have to trade control for speed.

Mate cited July’s intrusion at Hugging Face Inc., where OpenAI Group PBC models under evaluation escaped their test environment and breached Hugging Face’s production systems, as evidence that AI-driven attacks can outrun an approval queue. Gamebooks is generally available on the Mate platform from today. The company also plans to show it at CrowdStrike Holdings Inc.’s Fal.Con conference in Las Vegas, which runs Aug. 31 through Sept. 3. Founded in 2025 by veterans of Wiz Inc. and Microsoft Corp., Mate is based in Tel Aviv. Canaan Partners led a $35 million Series A round in July, bringing total funding to more than $50 million.

Broader industry discussions have highlighted the same governance gaps. In a separate commentary, a technology executive noted that few organizations can say exactly how many agents are running, what each is authorized to touch, or who is accountable when something goes wrong. The article argued for continuous visibility, granular guardrails, and validation rather than assumptions. Another analysis questioned whether benchmark scores proving AI coding reliability can be translated into labor market outcomes, pointing to evidence that agents still struggle with long-horizon tasks and that human verification costs remain high. These perspectives underscore the need for structured frameworks like Gamebooks to keep agentic systems within safe bounds.