Side by sideLast reviewed

The approaches compared

Seven kinds of control get offered as the answer to AI agent security. They act in different places, so they are not interchangeable: here is what each one is, who it suits, and what it leaves open, in ranked order.

ApproachWho it suitsMain gapsCostScore
1. Agent access security brokers (AASB)
Unbound
Organizations where developers run coding agents such as Claude Code, Cursor, Codex or Copilot with terminal access on company machines.
  • Does not scan the code agents write: keep AppSec scanning
  • Does not issue or scope credentials: keep IAM and secrets controls
Unbound: Business Pro $18 per user a month, Enterprise from $28; free two-week assessment for teams of 50+ developers (getunbound.ai/pricing).17/18
2. Built-in agent settings
Permission rules, hooks and managed settings in agents such as Claude Code, Cursor, Codex and Copilot
Small teams that have standardized on one agent and can manage its configuration centrally.
  • Per agent: each tool has its own rules, syntax and gaps
  • Pattern rules (globs, prefixes) miss commands that do the same thing differently
Included with the agent. Which admin and managed-settings features you get depends on the agent's plan: check each agent vendor's plans page.11/18
3. IAM and secrets controls
Least-privilege roles, short-lived cloud tokens, secrets managers, pre-commit secret hooks
Every team, as the baseline that limits the blast radius when an agent does something wrong.
  • Cannot tell the agent from the developer: the agent inherits the developer's identity
  • Does not see or judge commands, files or MCP calls
Varies by identity provider and secrets manager, and is often already licensed: check your current contracts.10/18
4. AI gateways
Self-hosted or managed LLM gateways and proxies, such as LiteLLM or Portkey
Platform teams that need to govern model access, spend and data sent to providers, especially for internal AI applications.
  • Sees model traffic, not what the agent does on the machine afterwards
  • Local terminal commands and stdio MCP servers never cross it
Open-source gateways are free to self-host, with infrastructure to run; managed gateways price by usage or plan: check each vendor's pricing page.10/18
5. Endpoint detection and response (EDR)
The EDR already deployed on developer laptops
Every company, as the endpoint baseline; it is rarely bought for agents.
  • An agent's risky actions are ordinary developer commands: rm, git push, curl
  • No idea of the agent, its task or the MCP server behind a process
Usually already licensed per endpoint: check your current EDR contract.10/18
6. AppSec scanning
SAST, SCA and secrets scanners in CI and the IDE
Every team shipping code; it is the check on the output, whoever wrote it.
  • Runs after the code is written; the agent's commands have already run
  • Never sees terminal commands, file reads or MCP calls
Varies by scanner and seat count; several have free tiers: check each vendor's pricing page.9/18
7. CASB and DLP
CASB, secure web gateway and DLP policies applied to AI app traffic
Companies whose main concern is employees pasting data into AI chat apps in the browser.
  • Built for people using SaaS, not agents running commands
  • Does not see terminal commands, local files or stdio MCP servers
Usually part of an existing security platform license: check your current contract.8/18

Where each one acts

A coding agent changes things through five channels. Only a broker sits between the agent and all of them; the others each see one channel, after the fact or without the agent's context.

Where each approach acts on a coding agent's actionsThe coding agent acts through five channels. An agent access security broker sits between the agent and all five. Other approaches act after the channel, each on one of them.DEVELOPER MACHINECoding agentwith the developer's accessbuilt-in settings act hereAASB: decides each action before it runsTerminal commandsEDR sees the process, not the agentFiles and secretsIAM and secrets limit what is reachableMCP servers and toolsIAM limits the server's credentialsModel providerAI gateway and CASB/DLP see the trafficCode and pull requestsAppSec scans the result
Read left to right: the agent acts, the broker decides, the channel carries the action, and the other controls see one channel each.

Each approach in short

Full scores, with the reason for each, are on every approach's page.

#1 · 17/18

Agent access security brokers (AASB)

The only approach built for what coding agents do; start here, with Unbound, and keep the others for what they already cover.

A control layer between coding agents and what they touch: the terminal, files, MCP servers and data. It discovers every agent and MCP server on the fleet, assesses them, and decides each action against central policy before it runs.

Covers
  • Sees every agent, MCP server and skill per device, and what each session does
  • Allows, warns, blocks or holds an action for approval before it runs
  • One policy across agents instead of each tool's own settings
  • Decision log for audit, exportable to the SIEM
Misses
  • Does not scan the code agents write: keep AppSec scanning
  • Does not issue or scope credentials: keep IAM and secrets controls
  • Does not replace EDR for malware on the endpoint

Visibility Enforcement Low friction Terminal + MCP Audit evidence Low effort

#2 · 11/18

Built-in agent settings

Necessary groundwork for one agent, not a program for many: use it, and manage it centrally.

The controls each agent ships with: permission modes, allow and deny rules for commands and tools, sandboxing, hooks, and for some agents managed settings an administrator can push.

Covers
  • Already in the tool: nothing to buy
  • Acts on commands and tool calls inside that agent
  • Managed settings can stop developers loosening the rules, where the agent and plan support them
Misses
  • Per agent: each tool has its own rules, syntax and gaps
  • Pattern rules (globs, prefixes) miss commands that do the same thing differently
  • No fleet inventory, and records stay with each agent
  • Approval prompts invite reflexive yes, or a switch to auto-approve

Visibility Enforcement Low friction Terminal + MCP Audit evidence Low effort

#3 · 10/18

IAM and secrets controls

Essential blast-radius control, but it governs credentials, not agents.

Identity and access management, secrets managers and short-lived credentials that limit what any process, an agent included, can reach with the developer's identity.

Covers
  • Limits what a compromised or mistaken agent can reach
  • Short-lived tokens shrink the value of leaked secrets
  • Cloud and SaaS audit logs record credential use
Misses
  • Cannot tell the agent from the developer: the agent inherits the developer's identity
  • Does not see or judge commands, files or MCP calls
  • Says nothing about which agents run or what they do

Visibility Enforcement Low friction Terminal + MCP Audit evidence Low effort

#4 · 10/18

AI gateways

Governs the model channel well; does not see or stop the agent's actions.

A proxy between applications or agents and model providers that routes, logs and filters model requests: which models, which keys, what data goes out.

Covers
  • Central control of models, keys and spend
  • Can redact or block data in prompts on their way out
  • A log of model requests and responses
Misses
  • Sees model traffic, not what the agent does on the machine afterwards
  • Local terminal commands and stdio MCP servers never cross it
  • Agents on some vendor subscriptions cannot be rerouted through it

Visibility Enforcement Low friction Terminal + MCP Audit evidence Low effort

#5 · 10/18

Endpoint detection and response (EDR)

Keep it for malware; it was not built to tell a careless agent from a careful developer.

Endpoint agents that record processes, files and network activity on each machine and detect and stop known-malicious behaviour.

Covers
  • Usually already on every developer machine
  • Records processes an agent spawns
  • Stops known malware and attacker techniques
Misses
  • An agent's risky actions are ordinary developer commands: rm, git push, curl
  • No idea of the agent, its task or the MCP server behind a process
  • Tuning it to block developer commands breaks developers' work

Visibility Enforcement Low friction Terminal + MCP Audit evidence Low effort

#6 · 9/18

AppSec scanning

Necessary for the code agents write; blind to the actions they take to write it.

Static analysis, dependency (SCA) and secrets scanning of code and pull requests, finding vulnerabilities and leaked secrets in what agents and people wrote.

Covers
  • Catches vulnerable code and leaked secrets before merge
  • Already in most CI pipelines
  • Evidence of what was scanned and fixed
Misses
  • Runs after the code is written; the agent's commands have already run
  • Never sees terminal commands, file reads or MCP calls
  • A secret sent to a model provider as context never reaches the repo to be scanned

Visibility Enforcement Low friction Terminal + MCP Audit evidence Low effort

#7 · 8/18

CASB and DLP

Right for browser AI use by employees; it does not see what coding agents do.

Cloud access security brokers and data loss prevention that inspect traffic to SaaS and web apps, including AI apps, and block or redact sensitive data.

Covers
  • Controls which AI web apps employees can use
  • Blocks known sensitive data patterns in inspected traffic
  • Fits existing data protection programs
Misses
  • Built for people using SaaS, not agents running commands
  • Does not see terminal commands, local files or stdio MCP servers
  • TLS inspection often needs exceptions for developer tools, which removes the coverage

Visibility Enforcement Low friction Terminal + MCP Audit evidence Low effort