What You Need to Know
Objective 2.1 gave you the reviewer's stance. This objective names the three failure types a review looks for. A hallucination is content that sounds convincing but is not grounded in fact. An inconsistency is an output that disagrees with itself. Bias is skewed framing, assumptions or language. Each has recognisable signs, and each needs a different response.
This lesson uses one made-up example throughout: a marketing agency. An account manager there gives Claude three jobs in one week. She wants the steps for the monthly hours report in the agency's time-tracking system, a campaign results summary with a table, and performance review notes for her team. Each job can fail in one of the three ways.
Hallucination: Fluent, Specific, Fabricated
The Help Center says Claude "can occasionally produce responses that are incorrect or misleading". It "can display quotes that may look authoritative or sound convincing, but are not grounded in fact". That last phrase is the working definition. A hallucination is not vague or hesitant. It appears as an exact-looking figure, a named source or a numbered list of steps. An invented detail has the same confidence as a real one, so specificity is not evidence. For any detail, ask "where would this have come from?" Asking "does this sound right?" tests only the confidence.
Claude can also hallucinate its own capabilities. A separate Help Center article warns that Claude may falsely claim to have sent an email, produced an external document or supplied a working link. Claude has no access to email or file transfer unless they have been explicitly integrated, which means connected to Claude. A claim that Claude has done something outside the conversation is a claim to check. It is not a confirmation that the delivery happened. If Claude says it has emailed the campaign figures to the client, the manager checks with the client.
Where the Risk Concentrates
Hallucination risk depends on grounding: how much material Claude has to work from. The risk is highest where Claude has least to work from:
- your organisation's private, internal systems and documents Claude was never given
- very recent events that happened after its training data (the material Claude learned from) was collected
- precise figures that were not in the source material
The highest-risk type of output is a set of confident numbered instructions for an internal tool whose documentation Claude has never seen. These instructions are fluent and specific, but nothing supports them. The hours-report steps are this type of output. Claude names menus in a time-tracking system that it was never shown, and some of those menus may not exist. The less grounding Claude could possibly have, the less its specificity means.
Inconsistency: The Output Disagrees With Itself
You need no external source to detect an inconsistency. One common form is a number in the prose that contradicts the table. Another is a recommendation that contradicts the findings. A third is two sections that give different dates for one event. An internal contradiction is a warning about the reliability of the whole output. It is not a single typo to fix. The process that produced one visible failure was just as able to produce invisible ones. So re-verify the whole output against its source, rather than fixing the one mismatch you noticed and sending the rest. If the campaign summary's prose and table give different click figures, the manager checks the whole summary against the campaign data.
Bias: The Pattern, Not the Sentence
Bias has the most serious consequences in outputs about people: recruitment material, performance reviews, customer descriptions, and anything that ranks or characterises groups. Anthropic's Usage Policy treats employment, housing, finance and healthcare decisions about individuals as high-risk uses. This shows how much is at risk. The skew is in the pattern, so compare the descriptions with each other rather than read sentence by sentence. Suppose the review notes describe older staff by reliability and younger staff by innovation. No single sentence is objectionable, but the output still is. Look for coded language (words that suggest a group without naming it) and unstated assumptions about age, gender, background or ability. Then apply the read-aloud test: would the framing be acceptable if someone read it aloud to the group that it describes?
In the exam
The exam tests whether you can distinguish the three failure types and respond to each correctly, because the correct responses differ. For inconsistency, the tested point is the response: re-verify the whole output. Wrong options match a failure type with an incorrect response. They fix one contradiction, delete one biased word, or ask Claude to confirm its own invented steps.
Match the response to the failure type
Verify a hallucination against an authoritative source. When you find an inconsistency, re-verify the whole output. Rework bias as a pattern across the document. Asking "is this right?" on its own confirms nothing. Anthropic documents two checks that use only the chat. The first is a re-run: send the same request again and compare the two outputs. A figure or step that is different in the two outputs can be a hallucination. The second is a follow-up that asks Claude to verify its previous statements and flag what does not fit together. Neither check confirms a fact. Each check gives you a list of items to check against the source. For the hours-report steps, that source is the time-tracking system's documentation. Objective 2.3 sets out what a self-check can and cannot do.
Exam Traps
Trusting confident, step-by-step instructions for a private internal system
Claude has no access to an organisation's internal software unless material about it was provided in the conversation, so specific detail with no source behind it is the highest-risk type of hallucination. Treat the steps as unverified and check the system's own documentation.
Believing Claude's claim to have sent an email or produced an external file
Claude can hallucinate its capabilities and has no access to tools that are not integrated, so a completion claim is a claim like any other. Verify that the action actually happened rather than treating the statement as confirmation.
Fixing the one contradiction you spotted and sending the rest
An internal inconsistency is evidence the output's process failed somewhere, so the whole output needs re-verification against its source. Fixing the one visible contradiction keeps a document that has already shown you it cannot be trusted line by line.
Treating bias as one bad word to delete
Bias is in the pattern across descriptions: who gets caveats, whose achievements get qualifiers, which groups are framed in different terms. Deleting a single biased phrase leaves the skew intact, so rework the pattern, not the word.
Practice Scenario
An office manager at a recruitment agency asks Claude how to produce the monthly placements report in the agency's candidate database software. Claude has had no access to this software or to anything written about it, yet it replies with confident, numbered steps naming specific menus and buttons. She still wants to use Claude for other work that involves this software. What should the manager do with these steps?
Build Exercise
Run a Three-Failure Sweep
What you'll learn
- How to separate traceable claims from unverified ones
- How to respond when an output contradicts itself
- How to read across descriptions of people to find a biased pattern
- How to match the correct response to each failure type
- Collect three real Claude outputs from your work: one making factual claims, one containing the same figures in more than one place, and one describing or comparing people
Why: Each output type is where one failure type most often appears, so the exercise teaches all three failure types on real material.
You should see: Three outputs open and ready: a claims document, a figures document, and a people document.
- Hallucination check: underline every specific that did not come from material you supplied. Specifics are figures, names, references and instructions. Mark each one traceable or unverified
Why: Hallucinations appear as specific details, so every specific without a source is a possible hallucination until traced.
You should see: A marked-up output where every specific is labelled one way or the other.
- Inconsistency check: compare every figure, date and conclusion that appears more than once, and where two appearances disagree, flag the whole output for re-verification
Why: The tested judgement is that a contradiction discredits the process, not just the sentence it appears in.
You should see: Either a cross-check with no mismatch or a flag on the whole output, never a single corrected number.
- Bias check: compare the descriptions of people with each other rather than read each one alone, and note whether different people or groups are systematically described in different terms
Why: No single sentence needs to be objectionable for the output to be; the pattern is the finding.
You should see: A note on whether the framing pattern passes the read-aloud test.
- For each flag you raised, write down the correct response: verify against a source, re-verify the whole output, or rework the pattern
Why: Naming the response helps you remember which response matches which failure type, and the exam tests this: each failure type has its own fix.
You should see: A three-line response plan, one line per failure type.
Sources
- Claude is providing incorrect or misleading responses. What's going on? — Anthropic
- Claude is producing links that don't work and falsely claiming that it has sent emails or produced external documents. What's going on? — Anthropic
- How up-to-date is Claude's training data? — Anthropic
- Reduce hallucinations — Anthropic
- Usage Policy — Anthropic
- Claude Certified Associate – Foundations Exam Guide — Anthropic