Contact Open demo
Report / Agent Reachability, 2026

AI agents are everywhere. Few teams know what they can reach.

Agents, MCP servers and non-human identities now hold real credentials. Companies know they have to understand that access, but many are still working out how to map it and measure it.

The Agent Reachability Report follows authority from the agent to the data: what exists, what it reaches, how access travels and what a security team fixes first.

What the report looks at

Reach matters more than inventory

Counting agents tells you little. The report follows which agents and identities can reach crown-jewel data, and through which route.

Authority travels

Roles, grants, tokens and MCP servers pass access along. A path is only as short as its weakest hand-over.

Fixes need proof

Each recommended fix is tied to the path it cuts, and the next scan shows whether it held.

Summary

Most security reporting counts assets. For AI agents that misses the point: an agent matters because of what it can reach, and an identity matters because of what it can pass on.

The report maps that authority, shows where it breaks, and states plainly what a map like this cannot tell you.

Inside the report
Chapter 01Who acts: agents, identities and keys

Agents, service accounts, keys, OAuth apps and MCP servers, and who owns them.

Chapter 02What they reach

Crown-jewel data stores and the identities that can get to them.

Chapter 03How authority travels

Roles, grants and tokens, hop by hop, from an agent to the data.

Chapter 04What gets fixed first

The changes that close the most exposure, in the order to do them.

Chapter 05Method and limits

How every figure was counted, and what a route cannot prove.

nerl0 ResearchThe Agent Reachability Report · 2026

Who acts: agents, identities and keys

Exposure over time
Evidence coverage over timeShare of each framework's mapped checks that passed, among those measured. Not an assessment of compliance; frameworks are not comparable with each other.
↗ Passing rate ⌄ScanDayWeek
0%10%20%30%40%50%60%11 Sep17 Sep23 Sep29 Sep05 Oct10 Oct16 Oct22 Oct28 Oct03 Nov14 Oct · scan 17NIS233%NIST CSF18%ISO 2700118%CIS6%
NowISO 2700121%NIST CSF24%NIS247%CIS8%
nerl0 ResearchThe Agent Reachability Report · 2026

What they reach

Coverage by area
Evidence coverage by areaShare of mapped checks that passed, among those measured.
HighMidLowNo mapped check
Area of the company
ISO 27001Annex A · 2022 · 10 of 93 mapped
NIST CSF2.0 · 4 of 106 mapped
CIS Controls8.1 · 2 of 153 mapped
NIS22022 · 6 of 42 mapped
Governance, policy & risk
–
50%1/2
–
67%2/3
People & HR
100%2/2
–
–
–
Assets & information
50%1/2
67%2/3
–
75%3/4
Identity & access
38%3/8 of 10
43%3/7 of 10
0%0/1 of 4
71%5/7
AI agents & non-human identities
–
–
100%1/1
33%2/6 of 12
Cryptography & secrets
–
–
–
0%0/1 of 2
nerl0 ResearchThe Agent Reachability Report · 2026

How authority travels

agenttokenrolecrown jewel
Method

Where the data comes from

Every figure is traced to a scan of connected platforms, and every chapter says where it stops.

Read-only scans

nerl0 reads roles, keys, grants and bots through read-only connectors. Nothing is installed on your hosts.

SourceCloud, identity, code and AI platforms
CountedAgents, identities, grants and the routes between them
PerScan, with the platform behind each number

What it cannot show

A route shows granted authority, not proof that the access was used. The report states this wherever it matters.

Not countedWhether a route was ever used
Not seenPlatforms that are not connected
StatedLimits are written next to each figure

Related resources

The Question Library

Every question behind the report, ready to activate in your own workspace.

Browse the questions →How Evidence works

See how nerl0 backs an answer with a scan instead of a claim.

Read about Evidence →Explore the demo workspace

Look at reach, paths and issues with demo data.

Open the demo →

Get the Agent Reachability Report free, in your inbox

How teams map, rank and cut what their AI agents can reach.

Book a demo