Agent Policies

87 design-time policies across 47 agents · 14 policy templates · last review 2 days ago

Tool-access

5

Standard Tool-access — BFSI Customer-facing

Read-mostly tools for customer-facing agents; write tools rate-limited.

Tool-accessEU AI ActDPDP

17 agents · v4.2reviewed 3d ago

Restricted Tool-access — High-risk Decisions

Hard allowlist; no write tools without HITL.

Tool-accessSR 11-7EU AI Act

6 agents · v3.1reviewed 6d ago

Read-only Tool-access — Investigation Agents

Investigation agents may read across systems but never write.

Tool-accessSR 11-7DPDP

8 agents · v2.8reviewed 1d ago

Custom: claims-investigation-agent tool policy

Per-agent override layered on Standard BFSI Customer-facing.

Tool-accessEU AI Act

1 agent · v1.7reviewed 4h ago

Custom: aml-investigation-agent tool policy

Restricts SAR-prep agent to AML-scoped data tools only.

Tool-accessSR 11-7DPDP

1 agent · v1.4reviewed 2d ago

Action-approval

5

Standard Approval Thresholds — Insurance

Payouts >$5K → adjuster; >$50K → manager.

Action-approvalEU AI Act

9 agents · v3.3reviewed 5d ago

Standard Approval Thresholds — Retail Banking

Refunds, fee-reversal and credit-line changes thresholds.

Action-approvalDPDP

11 agents · v2.9reviewed 4d ago

Strict Approval Thresholds — Mortgage

All disclosure & eligibility outputs require named officer sign-off.

Action-approvalSR 11-7EU AI Act

4 agents · v4.1reviewed 2d ago

Multi-approver — Frontier Decisions

Two-of-three approvers for novel-decision categories.

Action-approvalSR 11-7

2 agents · v1.2reviewed 1d ago

Custom: payout-recommendation approval

Routes payout recommendations >$5,000 to claim-team queue.

Action-approvalEU AI Act

1 agent · v1.6reviewed 4h ago

Memory

5

Standard PII-redaction Memory

Auto-redacts Aadhaar, PAN, account #, name+DoB combos before persistence.

MemoryDPDPEU AI Act

22 agents · v5.0reviewed 6d ago

90-day Retention — Customer-facing

Long-term memory aged out 90d after last touch.

14 agents · v2.4reviewed 7d ago

30-day Retention — Internal-tooling

Shorter retention for internal-only agents.

10 agents · v2.1reviewed 2d ago

No-memory — High-sensitivity

Stateless enforcement for high-sensitivity workflows.

MemorySR 11-7DPDP

3 agents · v1.5reviewed 3d ago

Long-term Memory with Encryption — Investigation

Investigation memory encrypted with KMS keys; team-scoped access.

MemorySR 11-7DPDP

6 agents · v2.0reviewed 1d ago

Inter-agent

3

Standard Inter-agent Communication

Sub-agent invocation only with explicit task scope; full audit logs.

18 agents · v2.2reviewed 4d ago

Restricted Inter-agent — Sandboxed

No cross-team agent invocations; sandbox per crew.

Inter-agentSR 11-7

4 agents · v1.3reviewed 1d ago

Audit-only Inter-agent — High-risk

Inter-agent messages logged + reviewed weekly.

Inter-agentEU AI Act

3 agents · v1.1reviewed 5d ago

Resource-budget

4

Standard BFSI Customer-facing Budget

Per-task: 25 tool calls · 50K tokens · $0.50 · 5 min.

Resource-budget

17 agents · v3.0reviewed 3d ago

Strict Internal-tooling Budget

Tighter budgets: 15 tool calls · 30K tokens · $0.20.

Resource-budget

10 agents · v2.5reviewed 5d ago

Investigation-agent Budget

Longer wall-clock (15min); higher tool-call ceiling (60).

Resource-budget

8 agents · v1.9reviewed 2d ago

Crew-agent Budget (multi-agent)

Crew-level budget with per-agent sub-budgets.

Resource-budget

5 agents · v1.7reviewed 4d ago

Escalation

3

Standard 3-failure Escalation

After 3 failed tool calls, escalate to human.

24 agents · v3.1reviewed 6d ago

Adversarial-pattern Auto-escalation

Detected jailbreak / prompt-injection auto-routes to security.

EscalationEU AI Act

17 agents · v2.4reviewed 1d ago

Cost-threshold Escalation

Per-task cost >$1.00 escalates for review.

14 agents · v1.8reviewed 3d ago

Rollback

3

Standard Rollback

Rollback to last checkpoint; completed actions preserved.

19 agents · v2.3reviewed 4d ago

Multi-approver Kill-switch

Two named operators required to invoke kill-switch.

RollbackSR 11-7

7 agents · v1.6reviewed 2d ago

Auto-pause on Critical Finding

Open critical red-team finding auto-pauses agent in production.

RollbackEU AI Act

11 agents · v1.2reviewed 1d ago