Backed by $12 million in seed funding, Metano helps enterprises secure AI agents wherever they operate, including endpoints, cloud and SaaS.
Metano Product Team
Kiro is Amazon’s agentic coding tool, available as an IDE and a CLI. Kiro Crew is an open-source workspace in the Kiro ecosystem that keeps agents running on your own hardware. You can access its agents from a desktop app, web dashboard, CLI, Slack, or Discord.
To keep the agents from affecting the system, Kiro Crew runs their tools inside an OS sandbox, which also prevents an agent from changing its own security settings. Every tool call is supposed to pass through Kiro Crew’s security gate. Metano’s research team found two issues:
Together, they let a prompt-injected agent turn the sandbox off with a single write to that file. Once the sandbox was off, the agent could rewrite the rest of its security settings too. The two compose into the cleanest version of the attack, but the config.json write does not need Agent Channels. Sensitive-path protection only ever covered the file-edit tool, so a shell command reaches the file on any surface, including the ones that do call the gate.
We reported both issues in September, and the Kiro team fixed them. The pull request that fixed the gate bypass sums up the cause in one sentence: “A check that each caller must remember is only as strong as the caller that forgets it.”
Kiro Crew’s sandbox uses Seatbelt on macOS, and user and mount namespaces on Linux. Besides isolating the agent’s tools, it makes the files that hold the agent’s security settings (the policy files) read-only, so a shell started by the agent cannot change them.
The sandbox itself can be switched off by setting agent.sandbox to off in config.json, one of those policy files. But config.json was not on the sandbox’s read-only list. In our tests, writes to the other policy files failed with a permission error, while the write to config.json succeeded.
Kiro Crew’s security gate should have refused the write to config.json. The gate (the codebase calls it the PreToolUse gate) runs three checks:
config.json and the other policy files.execute_bash for one agent or keeping it in one folder.The first two checks run even when the feature asking has already auto-approved the call, so that automatic approval can never switch them off. But this design works only if every feature that can approve a tool call actually goes through the gate.
Agent Channels did not. A channel is a room in the dashboard where several agents, each with a role, work on one problem in a shared transcript. Anything that changes state is supposed to come back to the user as an approval card. When a model asks to run a tool, the channel code takes one of four branches, and none of them calls the security gate. In pseudocode:
# How Agent Channels answers a tool requestif the tool is a messaging or session tool: rejectelif YOLO (approve everything) is on or the channel is trusted: approve # no checkelif a command grant matches:
if the binary behind the name is still the approved one:
approve # no security-gate check
else:
fall through to the approval card
else: # ask the user (approval card)
pin the binary behind the approved command
if the user clicks "trust":
How each feature handles tool approvals.
The first branch is a short list of blocked tools covering only messaging and session tools, not shell or file tools. The command-grant branch does check something: whether the program behind the name is still the file the human approved, refusing a name that resolves through an agent-writable directory or whose binary was swapped since. It never asks whether the command is allowed, so a grant for a command the operator’s denylist forbids is still approved.
One “trust” click covers every later call. Clicking “trust” on an approval card does more than approve one call. It marks the whole channel as trusted. After that, every tool call in the channel takes the second branch and is approved with no check, including calls the user never sees.
In a trusted channel, this meant:
execute_bash had no effect, because the code that enforces it never ran.config.json was approved, even though sensitive-path protection exists to block exactly this.Put together, the two issues let a single channel message turn off the agent’s sandbox:
config.json to turn off the sandbox.config.json is not read-only inside the sandbox.agent.sandbox, sees off, and does not create a sandbox.
From a channel message to an unsandboxed agent, in 5 steps. Each step removes the protection that would have stopped the next one.
We reproduced this on macOS with the agent in the default auto mode, driving the real sandbox wrapper directly against a throwaway crew home rather than a live agent. A shell inside the sandbox tried to write to the policy files, first with the sandbox on and then again after the config.json write had turned it off. Nothing else changed between the two runs, so the permission errors in the first run came from the sandbox.
# sandbox on (default auto mode)write config.json -> rc=0 OKwrite denied_commands.json -> rc=1 blockedwrite security_policy.json -> rc=1 blocked
# after the config.json write, the sandbox mode is offwrite denied_commands.json -> rc=0 OKwrite security_policy.json -> rc=0 OKThese files matter. With them writable, the agent can turn off rules in the command denylist and rewrite its governance rules. Two more effects come from reading the code: the agent could enable computer use for itself, and files the sandbox normally hides, such as .env files with Slack and Discord tokens, would become readable.
This is not zero-click, and it is not unauthenticated remote code execution. Four conditions must be satisfied to reach the gate bypass. The config.json write is reachable without them.
kirocrew app enable channels. This limits who was affected, but not the impact for workspaces that turned it on.config.json write, and models do not follow injected instructions every time.The fourth condition sets the severity. When it holds, no human is involved, and Amazon’s security team agreed to rate this High. When it does not, the call goes to the approval card, which still skips the gate but needs a person to click. Amazon rated that Medium.
What we tested. We ran the real channel handler on the trusted path and saw that the security check never ran. The other approval paths come from reading the code. We reproduced the sandbox shutdown on macOS, and for Linux we read the code but did not run it. Also, off only turns off Kiro Crew’s own sandbox. Any sandbox in the backend CLI is separate, and our test did not measure it.
If you run Kiro Crew:
src/kiro_crew/permission_floor.py exists. If it does not, the build predates PR #14132.sandbox.min_level governance rule so a config file cannot lower the sandbox mode.Channel gate bypass
SECURITY.md asks.config.json sandbox setting
config.json and config.local.json read-only.At the time of publication, neither issue has a CVE or GHSA. We will update this post if one is issued.

Harsh Akshit
AI Researcher
Harsh Akshit is an AI and Application Security Researcher specializing in AI security and Red Teaming. He currently focuses on exploring security risks related to agentic AI systems at Metano.
The latest news, technologies, and resources from our team.
Backed by $12 million in seed funding, Metano helps enterprises secure AI agents wherever they operate, including endpoints, cloud and SaaS.
Metano Product Team
A compromised AI tool could slip an ordinary-looking link into Jan’s chat. One click handed an attacker page the app’s internal powers, and it ran any program on the user’s machine, no password, no second prompt.
Parth Shukla
Since 21 October, computers infected with Mini Shai-Hulud have uploaded stolen secrets to 990 public GitHub repositories; the newest was created just 12 minutes before we counted. They upload using a stolen GitHub token that infected computers pass among themselves.
Japneet Singh
Anthropic’s inference hooks let a security server allow or deny every Claude Enterprise prompt and tool call before the model runs. Metano is now that server. It works on any device, with nothing installed.
Metano Product Team
Exposed LiteLLM gateways that still use the sample admin key will hand out API keys to anyone who asks. Each key is unique, so we could follow it to see who used it next.
Neelima Rustagi and Sonali Tyagi
An agent skill runs with everything the agent that loads it can do. New research from Metano MARS Labs, Purdue University and VulNow profiles more than 6,000 skills sourced from a couple of different marketplaces, to measure how much of that authority they put into play.
Japneet Singh
Occasional updates, new content, and insights. No spam; unsubscribe anytime.