Agent blast-radius scanner for cloud-hosted fintech agents
Lists every credential, metadata endpoint, peer agent and workplace identity each deployed agent can reach, then shows which public agent could take over the rest.
Crowded
Written for: Small startup team
Next step
Call the head of cloud security at a fintech running agents on Bedrock AgentCore. Ask whether, on the day of the Zenity disclosure, they could list every credential each of their agents could reach.
Why now
The AWS bug is patched, but the pattern remains: each agent has a separate identity, overly broad default permissions and access to its neighbours. Lab-reported incidents of agents escaping their test scope, and the ARTEX attack agent used against Korean banks, make this a board-level question for fintech security teams.
What became possible
Zenity showed that one public agent on Bedrock AgentCore could read, rewrite and delete every other agent in the same AWS account and region by reaching an internal credential interface. Agents are now getting their own identities: Google's Gemini agents get email addresses, and Bops gives bots their own computer, email and phone number.
Two ways in
- Agent reachability graph
Read-only scans of AWS, Google Cloud and Azure agent deployments that map what each agent can reach, highlight public-facing agents that could reach sensitive ones, and suggest least-privilege policies.
- Customer
- Cloud security leads at Series B to D fintechs running agents in production
- Model
- SaaS priced per agent scanned
- Agent identity inventory
Discovers agent-owned email accounts, API keys and phone numbers across workspace tools and checks them against an approved register.
- Customer
- Security teams at banks piloting Gemini Enterprise agents
- Model
- Annual licence
Competition
Both product ideas are already sold by well-funded vendors. The reachability graph idea overlaps with Wiz AI-SPM and Microsoft Defender CSPM: both already show agent blast radius and attack paths across AWS, Azure and Vertex.
Full competition notes (11 competitors)
Both product ideas are already sold by well-funded vendors. The reachability graph idea overlaps with Wiz AI-SPM and Microsoft Defender CSPM: both already show agent blast radius and attack paths across AWS, Azure and Vertex. Microsoft currently bundles its version for free. The identity inventory idea overlaps with Token, Astrix, SailPoint, Island and SPIRL Ledger, and Google's own Agent Registry gives each Gemini Enterprise agent a SPIFFE identity. Selling a general-purpose "agent graph" or "agent inventory" to a fintech that already pays for Wiz or Defender will be very hard. The openings I can see are narrower and are inferences from what the sources don't mention, not confirmed gaps. First, the sources describe the big platforms modeling permissions, not testing them. A new tool could check from inside the agent runtime what is actually reachable: instance metadata endpoints (IMDS), network paths, and peer agent-to-agent (A2A) or MCP delegation chains. That would show which internet-facing agent could take over others through chains of agents passing tasks to each other. Pillar's SAIL framework names this transitive reachability as a risk nobody maps. Second, none of the identity vendors mention agent-owned phone numbers or email inboxes inside workspace tools. Third, none of the sources show fintech-specific packaging, such as evidence mapped to DORA, FFIEC, NYDFS Part 500 or bank model-risk rules, or least-privilege policy fixes ready to hand to auditors. A newcomer would probably win only as a deep specialist on one of these angles, or by plugging into Wiz or Defender, not by competing as another posture-management platform.
- Wiz AI-SPM
Takes inventory of agents with their models, tools, MCP components and data sources, and states each agent's blast radius. It links each agent to the identities, workloads and data it can touch, and shows attack paths an exploited agent could take. Covers Bedrock, Vertex AI and OpenAI. This is the closest direct overlap with the reachability graph idea, and it is already in the hands of most cloud security leads.
- Microsoft Defender for Cloud (Defender CSPM AI security posture)
Finds AI agents across Azure, AWS and GCP Vertex AI (in preview), builds an AI bill of materials and runs attack path analysis. Currently bundled into Defender CSPM at no extra charge, which pushes pricing down for a standalone product.
- Zenity
Agent security platform whose Surface module covers discovery and blast radius. It also controls agent actions at runtime. Lists Gemini among the agent platforms it supports. Source is a third-party directory listing.
- Token Security
Non-human identity security vendor with an AI Discovery Engine. It combines identity, cloud, SaaS and AI platform data into a live inventory of agents and their access, and says which agents use which credentials. Overlaps heavily with the agent identity inventory idea.
- Astrix Security
Its AI agent security platform discovers managed and shadow agents, including on Google Vertex. Uses NHI fingerprinting of OAuth apps, service accounts, API keys and PATs to catch agents that were never registered. Directly covers checking agents against an approved register.
- Ledger (SPIRL) Gemini Enterprise integration
Takes inventory of Gemini Enterprise Agent Platform resources, agent identities, integrations and credential metadata, and flags secrets that need fixing. This aims at the same Gemini Enterprise bank pilot buyer.
- Google Gemini Enterprise Agent Registry and Agent Gateway
Built-in registry where admins see each agent's identity: a SPIFFE ID if the publisher exposed it, otherwise a fallback registry ID. Pairs with least-privilege IAM and a central policy gateway. This partly replaces a third-party approved-agent register.
- Island
Pulls non-human and agent identities from cloud, SaaS, repositories and AI tools. Also reads endpoint configs, including MCP configs, to find hardcoded credentials.
- SailPoint Agent Identity Security
Identity governance product that inventories AI agents, links them to their human owners and secures their credentials. Already sold into banks through existing IGA relationships.
- Backslash Agentic Fabric Graph
Graph of AI agents, MCP servers, models, skills and workstations, prioritized by blast radius. Focused on endpoints rather than cloud deployments.
- Gravitee
Publishes material on agent graph governance and attack path graphs. It models agents, service accounts, APIs and MCP servers as nodes connected by authorization and tool-call edges, and approaches the problem from its API and agent gateway products.
Checked 2026-10-09
Risks
- Cloud-security incumbents (Wiz, Palo Alto) and Zenity itself are moving here quickly.
- Cloud providers keep tightening defaults, which shrinks the most obvious findings.
- Security buyers prefer established vendors over a 3-person team.
Best on tech readiness and market pull; open space is the weak point (5).
tech 7 · pull 7 · space 5 · fit 6 · build 7
Exposed
Companies that could benefit or be threatened. News and context, not investment advice.
Momentum of its main topic
Sources
- One public-facing AI agent on AWS could read, rewrite, and delete every other agent in the regionThe Decoder
- From Reactive Containment to Proactive Assurance: Lessons from OpenAI, Anthropic, and Google Agent Security IncidentsarXiv
- Google brings agentic AI to Gemini, starting with businessesTechCrunch AI
- OrgoAI/bopsGitHub
- The Chinese developer of ARTEX says it has converted the AI agent into a closed-source project after CrowdStrike said it was used to hack South Korean banks (Brenda Goh/Reuters)Techmeme
What happened next
- +7d · 16 Oct 2026 nothing yet. No products matching the agent blast-radius scanner opportunity were found to have launched, been funded, or received significant press coverage between October 9 and October 16, 2026. While several agent security tools emerged earlier in 2026 (Backslash Agentic Fabric Graph in July, Salt Security in March, NewCore in June), none fall within the post-flag window.
Status history
- 2026-10-09 Flagged as buildable