Safety Settings

Configure tenant-wide guardrails for inbound threat classification and sending behavior.

Safety settings control how your tenant handles inbound threats, spam classification, and agent-safety guardrails. These are global settings that apply to all agents and mailboxes in your tenant.

Changes take effect immediately. Safety settings work alongside (not as a replacement for) policy engine rules.

View current settings

Via the CLI

CLI

molted safety get

Response

{
  "tenantId": "tenant_abc123",
  "quarantineHighInjection": true,
  "holdCriticalAnomalies": true,
  "blockCanaryViolations": true,
  "spamAction": "quarantine",
  "phishingAction": "quarantine",
  "malwareAction": "reject",
  "abuseAction": "quarantine",
  "impersonationAction": "quarantine",
  "spamThreshold": 0.5,
  "maxLinksThreshold": 5,
  "blockNoAuth": false,
  "blockedKeywords": [],
  "allowedSenders": [],
  "spamActionLowConfidence": "deliver"
}

Via the API

GET https://api.molted.email/v1/agent/config/safety-settings

Requires Bearer authentication.

curl

curl https://api.molted.email/v1/agent/config/safety-settings?tenantId=TENANT_ID \
  -H "Authorization: Bearer YOUR_API_KEY"

Update settings

Updates are partial -- only fields you include are changed. Unspecified fields keep their current values.

Via the CLI

CLI

molted safety update --settings '{
  "spamAction": "reject",
  "spamThreshold": 0.7,
  "blockedKeywords": ["unsubscribe me", "remove me"]
}'

Flag	Type	Required	Description
`--settings`	`json`	Yes	Safety settings as a JSON string.

Via the API

PUT https://api.molted.email/v1/agent/config/safety-settings

Requires Bearer authentication.

curl

curl -X PUT https://api.molted.email/v1/agent/config/safety-settings?tenantId=TENANT_ID \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "spamAction": "reject",
    "spamThreshold": 0.7,
    "blockedKeywords": ["unsubscribe me", "remove me"]
  }'

Settings reference

Threat response actions

These control what happens when a threat is detected. Valid values: "deliver", "quarantine", "reject".

Field	Type	Default	Description
`spamAction`	`string`	`"quarantine"`	Action for detected spam.
`phishingAction`	`string`	`"quarantine"`	Action for detected phishing.
`malwareAction`	`string`	`"reject"`	Action for detected malware.
`abuseAction`	`string`	`"quarantine"`	Action for abuse or harassment.
`impersonationAction`	`string`	`"quarantine"`	Action for impersonation attempts.
`spamActionLowConfidence`	`string`	`"deliver"`	Action when spam confidence is below the threshold.

Action meanings:

deliver -- the message is delivered to the mailbox normally.
quarantine -- the message is held in the approval queue for human review.
reject -- the message is silently dropped and not delivered.

Classification thresholds

Field	Type	Default	Range	Description
`spamThreshold`	`number`	`0.5`	0.1 - 1.0	Spam score cutoff. Lower values are more aggressive.
`maxLinksThreshold`	`number`	`5`	1 - 100	Maximum links per message before flagging as suspicious.

Guardrail flags

Boolean flags that enable or disable specific safety guardrails.

Field	Type	Default	Description
`quarantineHighInjection`	`boolean`	`true`	Quarantine messages with high prompt-injection risk.
`holdCriticalAnomalies`	`boolean`	`true`	Hold messages flagged as critical anomalies for review.
`blockCanaryViolations`	`boolean`	`true`	Block messages that trip canary tokens.
`blockNoAuth`	`boolean`	`false`	Apply a heavy spam penalty (+0.5) to messages failing all auth checks (SPF, DKIM, DMARC).

Allow/deny lists

Field	Type	Default	Max entries	Description
`blockedKeywords`	`string[]`	`[]`	100	Keywords that trigger quarantine. Matched case-insensitively against subject and body.
`allowedSenders`	`string[]`	`[]`	100	Email addresses or domains whose mail, when the message's DMARC result is `pass`, has a `spam` verdict downgraded to `clean`. Never bypasses phishing, malware, abuse, impersonation, or prompt-injection detection -- and a sender that matches without a passing DMARC is treated as not allowed at all. This is a manually-edited setting; the separate, auto-growing sender allow-list (`GET`/`DELETE /v1/agent/override/allowlist`) works the same way but is built from repeated not-spam marks, not typed in here.

Examples

Lock down a high-security mailbox

Block unauthenticated email, reject malware and phishing, quarantine everything else:

CLI

molted safety update --settings '{
  "malwareAction": "reject",
  "phishingAction": "reject",
  "spamAction": "quarantine",
  "blockNoAuth": true,
  "quarantineHighInjection": true,
  "spamThreshold": 0.3
}'

Relax settings for a trusted environment

When your mailbox only receives email from known partners, a person can relax the settings. From an agent key this returns 403 human_decision_required with a decisionUrl (see Who can change what):

CLI

molted safety update --settings '{
  "spamAction": "deliver",
  "spamActionLowConfidence": "deliver",
  "spamThreshold": 0.9,
  "allowedSenders": ["partner.com", "ops@internal.co"]
}'

Add keyword blocklist

Block messages containing specific phrases:

CLI

molted safety update --settings '{
  "blockedKeywords": ["crypto opportunity", "wire transfer", "urgent action required"]
}'

Note that blockedKeywords replaces the entire list on each update. Include all keywords you want active, not just new additions.

Who can change what

Agents may tighten these settings, never loosen them. From an agent key, a change that loosens anything returns 403 human_decision_required with a decisionUrl and a loosens list (or, while your account is in warn, goes through with a Molted-Deprecation: human-gate header). A person signed in to the portal can change anything: the spam filter rules are under Settings > Safety, where each pending change says whether it tightens or loosens.

Setting	Tightening (agents may)	Loosening (needs a human)
`quarantineHighInjection`, `holdCriticalAnomalies`, `blockCanaryViolations`, `blockNoAuth`	`true`	`false`
Threat actions (`spamAction`, `phishingAction`, `malwareAction`, `abuseAction`, `impersonationAction`, `spamActionLowConfidence`)	A stricter action (from `deliver` to `quarantine` to `reject`)	A weaker action
`spamThreshold`, `maxLinksThreshold`	Lower	Higher
`blockedKeywords`	Add entries	Remove entries
`allowedSenders`	Remove entries	Add entries

The same rules apply to PUT /v1/spam-filter/rules. See Loosening a control.

Errors

Error	Cause	Fix
`400 Bad Request`	Invalid action value or threshold out of range.	Use `"deliver"`, `"quarantine"`, or `"reject"` for actions. Keep `spamThreshold` between 0.1 and 1.0, `maxLinksThreshold` between 1 and 100.
`400 Bad Request`	Too many entries in `blockedKeywords` or `allowedSenders`.	Each list supports a maximum of 100 entries.
`401 Unauthorized`	Missing or invalid API key.	Run `molted auth init` to authenticate.

Inbound Email -- safety classification verdicts and spam queue management
Approval Queues -- review quarantined messages
Autonomy Levels -- per-mailbox outbound approval controls
Policy Simulation -- test policy engine rules before sending

On this page