Force Safe Alternatives Before AI Agents Run: The FORCE Decision

Why FORCE exists

Blocking a dangerous tool call is necessary, but it is not always enough. An AI agent still has a goal to complete, and a blunt denial can leave it guessing at the next move.

FORCE is the decision for that middle ground. It refuses the original risky action and returns a safe substitute or required safer path for the agent to use next.

What FORCE returns

A FORCE rule carries a SUBSTITUTE value alongside the normal rule message:

rule prefer-reviewed-deploy {
  enabled true
  priority 90
  severity error
  FORCE execution
  IF command CONTAINS "deploy production"
  SUBSTITUTE "open the deployment checklist and request approval"
  MESSAGE "Production deploys must follow the reviewed deployment path."
}

The original action does not run. The agent receives the reason and the substitute path, then can issue a new compliant tool call.

How FORCE differs from the other decisions

SSG decision summaries should name all six decisions:

Decision	Runtime meaning
ALLOW	Permit the tool call
DENY	Runtime block; refuse the tool call
ASK	Pause and require human approval
FORCE	Refuse the risky original and return a safe substitute
LOG	Permit and record for audit
SHADOW	Permit silently and record for monitoring

The important distinction is that FORCE is not an allow with nicer wording. It is a refusal of the original action, plus a deterministic safer path.

Where FORCE helps

FORCE works best for patterns where there is a known safer workflow:

• Prefer an audited file-read tool over shelling out to print a file.
• Route production deploys through a checklist instead of a raw command.
• Require scoped package-manager commands instead of broad global installs.
• Redirect unknown network fetches to a documented allowlisted source.

Each of those cases gives the agent useful next-step information while keeping the risky original action out of execution.

Audit and review

FORCE decisions are written to the same audit trail as ALLOW, DENY, ASK, LOG, and SHADOW. Reviewers can see the attempted action, matched rule, message, substitute, latency, and session context.

That makes FORCE useful for both live safety and later compliance review: the log shows not just what was refused, but the approved safer path SSG returned.

Get started

curl -fsSL https://install.sigmashake.com | sh -s -- -y
ssg init
ssg serve

Then add a FORCE rule with a SUBSTITUTE path for any repeated risky workflow where your team already knows the safer way to proceed.