Bay's $50M Agent Control Platform Exposes the Runtime Oversight Gap
What happened
Bay was founded in February 2026 by three alumni of Israeli Military Intelligence Unit 8200. It announced $50 million in total funding: a $37 million Series A led by Battery Ventures and a previously undisclosed $13 million seed round led by YL Ventures and Striker Venture Partners, as reported by SecurityWeek. The platform deploys without installing software on individual machines and claims full deployment across an enterprise agent fleet in under five minutes. Once running, Bay's automated inventory component discovers AI agents, coding assistants, tool servers (the connectors that let agents reach external systems), browser extensions, and local models. It then maps how they interconnect and flags risky tool combinations. Its Enforcer component evaluates every action an agent proposes during a session. It decides in milliseconds whether to allow it, refer it to a human, redirect it under policy, or block it entirely. Policies are generated and maintained automatically by a component called Lighthouse. Lighthouse analyzes 30 days of historical activity logs to derive what normal agent behavior looks like. The platform targets two root causes of harmful agent behavior identified in recent incident data: deliberate manipulation through hidden instructions inserted into content an agent reads (prompt injection), and unplanned drift in what an agent decides to do on its own (purpose drift). Both are documented in 1,000+ AI Rules, One Audit Gap: Runtime Evidence Is What Regulators Want and directly relevant to the Five Eyes Guidance on the Careful Adoption of Agentic AI Services.
Why it matters
- ·Regulators and auditors are increasingly asking for runtime evidence that AI agents operated within defined boundaries. Without a persistent inventory and action log, enterprises cannot demonstrate compliance with the EU AI Act (Regulation (EU) 2024/1689). They also cannot respond to examiner requests of the kind documented in Banking Examiners Are Now Asking Banks to Prove AI Agents Can Be Audited.
- ·Agents inherit the permissions of the user or service account they run under, meaning a single misconfigured or manipulated agent can access any system that account can reach. The control gap Bay targets is the same one behind incidents such as AI Agent Attack Wiped 100 Azure Storage Accounts in Seven Minutes. In that case, unconstrained agent access produced irreversible damage at speed.
- ·The emergence of a funded vendor category dedicated solely to agent runtime enforcement signals that existing controls are not sufficient on their own. Those controls include model guardrails, static policies, and manual review. Compliance programs that have not yet assessed runtime action monitoring now face a documented and commercially recognized control gap.
Stay ahead of stories like this
Get every Global AI governance development like this one, plus the rest of the week's developments. Every Thursday.
Recent issues
- State-linked attackers and criminal operators are actively weaponizing AI tools as entry points into enterprise systems, turning internal platforms and open-source frameworks into live attack vectors.5 Nov
- Three separate incidents this week each created a new compliance action item: verify your AI vendors' data sourcing, demand agent monitoring coverage guarantees, and audit trusted app identities in your cloud environment.29 Oct
Free every Thursday. Unsubscribe anytime.
Governance controls affected
What to do now
- ☐Commission a full inventory of every AI agent, coding assistant, browser extension, and tool connector currently running in your environment, including those deployed by individual business units without central IT approval.
- ☐Ask your security or engineering team whether any AI agent can take actions — such as sending emails, modifying files, or calling external services — without a human reviewing and approving each step before it happens.
- ☐Review audit trail completeness for agent sessions: confirm that logs capture what an agent was asked to do, what actions it proposed, and what it actually executed, not just the final output.
- ☐Assess whether your current agent deployments have controls that can detect and stop purpose drift — that is, an agent gradually doing things outside its original intended scope — not just outright policy violations.
- ☐Evaluate whether vendor contracts for AI agent platforms require the vendor to notify you when an agent takes an action outside its defined scope or triggers an anomaly, and update contracts where this obligation is absent.
What to watch next
Regulatory pressure on agent runtime controls is intensifying across multiple jurisdictions. The ICO has opened a formal agentic AI data protection probe. The Five Eyes Guidance on the Careful Adoption of Agentic AI Services names runtime enforcement and least-privilege access as baseline expectations. Compliance teams should monitor the EU AI Office's inspection program. It currently targets hiring, credit, and healthcare AI. Teams should watch whether it expands audit evidence expectations to include agentic runtime logs. The emergence of funded vendors in this space also signals that standards bodies and procurement frameworks may soon reference runtime agent monitoring as a named control requirement.