Anthropic interview questions

Showing 1–30 of 225 results

• A Conflict About ValuesMedium

  This question tests your ability to navigate professional disagreements where core values (e.g., speed vs. safety, or business goals vs. technical debt) are at odds, requiring a diplomatic yet principled resolution.

  AnthropicSoftware EngineerBehavioral & LeadershipOnsite
• A Disagreement That Changed Your MindMedium

  This question evaluates intellectual humility, evidence-based reasoning, and your ability to update your technical beliefs when presented with compelling counterarguments or data.

  AnthropicSoftware EngineerBehavioral & LeadershipOnsite
• AI Risk Values Round: Delaying a Risky Breakthrough, Worthless Equity, PersuasionMedium

  A culture round at a safety-focused AI lab: whether you would delay a risky capability breakthrough even if competitors would not, what you would do if your equity became worthless, a time you were persuaded, and how you repaired a damaged working relationship. It tests reasoning about AI risk, honesty and self-awareness.

  AnthropicSoftware EngineerBehavioral & LeadershipOnsite
• AI Safety Culture Round: Risk Views, Personal Ethics, and Changing Your MindMedium

  A senior software engineer culture interview at an AI company that probes your honest views on AI safety risks, how your personal moral principles hold up when acting on them has a cost, and whether you can genuinely change your mind. It tests candor over sycophancy, calibrated reasoning, and composure when the interviewer challenges your position.

  AnthropicSoftware EngineerBehavioral & LeadershipOnsite
• Account Balance with Expiring GrantsMedium

  Design a data structure that tracks time-limited credit grants, prioritizes consuming grants that expire earliest, and efficiently manages expired funds.

  AnthropicSoftware EngineerCoding & AlgorithmsOnsite
• Answer AI Safety Behavioral PromptsMedium

  This question evaluates behavioral and leadership competencies for a software engineering role, including alignment with organizational values, ethical judgment on AI safety, communication, mentorship, and influence on project and product decisions within the Behavioral & Leadership domain.

  AnthropicSoftware EngineerBehavioral & LeadershipOnsite
• Answer Culture and Project QuestionsMedium

  This question evaluates a software engineer's behavioral competencies—ownership, judgment, communication, and values alignment—by prompting story-driven examples about past projects, difficult tradeoffs, and motivation for joining an AI safety–oriented research organization.

  AnthropicSoftware EngineerBehavioral & LeadershipOnsite
• Answer culture-fit reflection questionsHard

  This question assesses a candidate's ability to demonstrate intellectual humility, accountability, and the capacity for self-directed growth in high-stakes technical environments.

  AnthropicSoftware EngineerBehavioral & LeadershipOnsite
• Answer general fit and AI safety questionsMedium

  Answer general fit and AI safety questions evaluates behavioral evidence, ownership, communication, trade-offs, and measurable outcomes in a realistic interview setting. A strong answer states assumptions, handles edge cases, explains trade-offs, and shows how to validate the result clearly.

  AnthropicSoftware EngineerBehavioral & LeadershipOnsite
• Apply bitwise tricks for performanceMedium

  This question evaluates proficiency with bitwise operations, branchless programming, low-level performance optimizations, and microarchitectural reasoning about pipeline stalls, instruction-level parallelism, and cache behavior.

  AnthropicSoftware EngineerCoding & AlgorithmsOnsite
• Assess an AI Research Organization and Explain Your FitHard

  Tests a candidate's understanding of frontier AI research labs, their ability to critically evaluate safety-driven operational decisions, and how well their engineering background aligns with safety-focused development.

  AnthropicSoftware EngineerBehavioral & LeadershipHR Screen
• Banking System SimulationMedium

  This question evaluates a candidate's ability to design a stateful, multi-level simulation system that processes a time-ordered stream of operations. It tests object-oriented design, incremental complexity management, and handling of deferred effects like scheduled cashback credits. Such object-modeling questions are commonly used in coding interviews to assess practical application of data structures under evolving requirements.

  AnthropicSoftware EngineerCoding & AlgorithmsOnsite
• Batch Image Resize and Rotate with Pillow: Exact Dimensions, Parallelism, Bad FilesMedium

  A three-part Python image processing exercise: resize and rotate images with Pillow while predicting the exact printed dimensions, speed up a large batch with multiprocessing and a staged pipeline, then keep the batch running through corrupt, oversized and unsupported files. It tests API precision, parallelism and robustness.

  AnthropicSoftware EngineerSoftware Engineering FundamentalsOnsite
• Batch-Apply Pillow Image Transforms, Then Parallelize: Processes vs ThreadsHard

  A Python coding exercise that applies a given sequence of Pillow transforms to a folder of cat images and then asks you to cut the total processing time. It tests clean per-image code, reasoning about CPU-bound work and the GIL, choosing between processes and threads, and sizing a worker pool.

  AnthropicSoftware EngineerSoftware Engineering FundamentalsOnsite
• Brainstorm How to Keep an AI Model Aligned With Its Intended GoalsMedium

  Anthropic Fellows final-round practice: a 15-minute open discussion on defining, detecting and training against misalignment and misuse, ending with one falsifiable experiment.

  AnthropicMachine Learning EngineerMachine LearningOnsite
• Build a Progressive Banking LedgerHard

  Design and implement an in-memory, timestamped banking ledger supporting atomic transfers, cashback, scheduled payments, account merges with alias resolution, and historical balance lookups.

  AnthropicSoftware EngineerCoding & AlgorithmsOnsite
• Build a Progressive Cloud Storage SystemHard

  Design and implement an in-memory cloud storage engine supporting user quotas, file operations, compression/restoration semantics, and glob-based search.

  AnthropicSoftware EngineerCoding & AlgorithmsOnsite
• Build a Progressive Employee TrackerHard

  Design an in-memory attendance and payroll tracking system that manages shift lifecycles, deferred rate promotions, overlapping bonus windows, and duration rankings.

  AnthropicSoftware EngineerCoding & AlgorithmsOnsite
• Build a Resize-and-Rotate Image Pipeline with MultiprocessingMedium

  Design a Pillow resize-and-rotate pipeline with explicit transform semantics, bounded multiprocessing, per-image failures, and safe output publication.

  AnthropicSoftware EngineerSoftware Engineering FundamentalsOnsite
• Build a concurrent web crawlerMedium

  This question tests your ability to design a concurrent crawler that manages shared state across multiple threads, handles URL de-duplication, and prevents infinite cycles during graph traversal.

  AnthropicSoftware EngineerCoding & AlgorithmsTechnical Screen
• Build a least-recently-used cacheHard

  This problem tests your ability to implement an efficient LRU cache using a combination of a hash map and a doubly linked list to achieve constant-time operations.

  AnthropicSoftware EngineerCoding & AlgorithmsOnsite
• Calculate Streaming Token Usage CostsHard

  Implement a robust token usage billing manager that accumulates streaming chunk deltas and non-streaming totals, enforces strict schema and sanity checks, and avoids floating-point precision loss.

  AnthropicSoftware EngineerCoding & AlgorithmsOnsite
• Compare Threads and Processes for Image ProcessingMedium

  Tests operating system concurrency fundamentals, memory ownership models, IPC serialization overhead, and empirical benchmarking strategies for CPU-bound image processing pipelines.

  AnthropicSoftware EngineerSoftware Engineering FundamentalsOnsite
• Compute Exclusive Time from Stack EventsHard

  Compute aggregate exclusive execution time for each function from nested start and end events, including recursive calls. Use a stack over a validated single-threaded trace, apply half-open time intervals, and return deterministic sorted results.

  AnthropicSoftware EngineerCoding & AlgorithmsOnsite
• Compute a Global Mode Across Distributed PartitionsHard

  Compute an exact distributed mode with local frequency combining, key-based reducers, global winner selection, communication accounting, and skew-aware recovery.

  AnthropicSoftware EngineerSoftware Engineering FundamentalsOnsite
• Compute an Exact Distributed Median with Limited CommunicationHard

  Compute an exact distributed median through global rank selection and count reductions, accounting for integer range, local search work, communication rounds, and even-size semantics.

  AnthropicSoftware EngineerSoftware Engineering FundamentalsOnsite
• Compute exclusive times and call stack from logsMedium

  This question evaluates a candidate's ability to process single-threaded execution logs and compute per-function exclusive durations while reconstructing the call stack at a given timestamp, testing stack simulation, interval reasoning, timestamp tie-handling, parsing of time-stamped events, and algorithmic complexity analysis.

  AnthropicSoftware EngineerCoding & AlgorithmsTechnical Screen
• Compute matrix trace and support updatesMedium

  This question evaluates proficiency in matrix manipulation, incremental updates, and complexity analysis, specifically maintaining main-diagonal and anti-diagonal sums under point updates.

  AnthropicSoftware EngineerCoding & AlgorithmsTechnical Screen
• Convert Sampled Call Stacks into Start and End EventsHard

  Convert sampled call stacks to inferred start/end events using ordered common prefixes, recursive frames, explicit final-open behavior, and last-N history limitations.

  AnthropicSoftware EngineerCoding & AlgorithmsOnsite
• Convert Sampled Call Stacks into Timestamped Function Start and End EventsMedium

  Convert a sampling profiler's timestamped call stacks into function start and end events for a trace viewer, explain why ends are emitted innermost first and starts outermost first, then redesign the converter for an unbounded stream. It tests stack diffing, recursion handling and streaming state.

  AnthropicSoftware EngineerSoftware Engineering FundamentalsOnsite