P5 warm-up · 8 questions

Warm up: Governance, Safety & Risk Management

A quick 8-question drill on CCAR-P P5, easy to hard. Rate your confidence as you go — a 5-minute win to find where you stand before a full practice set.

0 of 8 answered

P5 · Q1Escalate on policy, not model self-confidence
A support agent can issue refunds. It returns a refund action for a $4,000 order with a self-reported confidence of 0.94. Which control most appropriately decides whether the action executes without human review?
How confident are you?
P5 · Q2Single-output-filter anti-pattern
A team's entire safety posture is one regex-based output filter applied to the final text. All of the following are plausible upgrades; which best addresses the core weakness of a single output filter?
How confident are you?
P5 · Q3Least-privilege for tools and blast-radius containmentSelect 2
An agent has one credential granting read/write to the entire production database plus outbound email. Which two changes most reduce blast radius while preserving the agent's job of drafting order-status replies?
How confident are you?
P5 · Q4Human oversight for irreversible actions
An ops agent can delete cloud resources to save cost. Which design most appropriately governs this capability?
How confident are you?
P5 · Q5Must-hold rules in deterministic code
A trading assistant must never place an order that exceeds a client's regulatory position limit. Engineers propose several ways to enforce it. Which is the most defensible?
How confident are you?
P5 · Q6Signal triage checkpointSort each into a bucket
Route each production signal to the most appropriate destination.
A slow, steady rise in average tool-latency that is still within SLO
A guardrail failing open under load, letting unfiltered outputs through
A newly enacted regulation that changes the retention obligation for stored transcripts
A single 200-OK health-check ping logged every minute as expected
Injection attempts spiking against the input filter, all currently blocked
How confident are you?
P5 · Q7Bias/fairness and disparate-impact review
An AI screens loan applications. Compliance asks how you would detect disparate impact. Which approach is most appropriate?
How confident are you?
P5 · Q8Residual-risk acceptance and control drift
A control register lists a nightly PII-redaction job as the control for a privacy obligation. The job silently stopped running six weeks ago but the register still shows it as active. All statements are partly true; what is the root cause of the audit finding?
How confident are you?

Full P5 practice (3 attempts)

Back to readiness check