Which Companies Let You Use AI in Interviews in 2026

There is no industry-wide rule about using AI in a technical interview. There are employer rules, and as of September 21, 2026 they fall into three incompatible regimes: employers that require AI use in a named round, employers that permit it inside a specific interview and grade how well you use it, and employers that prohibit it and state disqualification as the consequence. The same candidate can hit all three in one week of onsite loops.

This article maps what named employers have actually said, with dates and sources, and then explains how to work out which regime applies to an interview that is not on any list. It is about employer policy, not platform settings — a topic the broader question of AI use during technical interviews covers from the process side.

The three regimes

Required: AI use is the assessment

Canva's engineering blog, published June 11, 2025, describes replacing its "Computer Science Fundamentals" screening round with an "AI-Assisted Coding" round for backend, machine learning, and frontend roles. Canva states that candidates are expected to use their preferred AI tools, naming Copilot, Cursor, and Claude, and that the problems are deliberately realistic product challenges rather than algorithm puzzles an assistant solves trivially (Canva Engineering Blog).

In this regime, declining to use AI is not a neutral choice. The round is designed around the assumption that you will, and the evaluation criteria Canva lists — knowing when to reach for AI, decomposing ambiguous requirements, catching and correcting AI-generated errors, getting a solution to production quality — are difficult to demonstrate if you write everything by hand.

Permitted in a named round, with the tool specified

Google is piloting a software engineering interview format in which candidates use Gemini during a code comprehension round. Brian Ong, Google's VP of recruiting, said the company is "rolling out a pilot for software engineering interviews to be more reflective of how our teams are operating in the AI era." The pilot was first reported from an internal document by Business Insider in May 2026, applies to select US-based teams hiring for junior and mid-career roles, begins in the second half of 2026, and has interviewers assess AI fluency including prompt engineering, output validation, and debugging (Entrepreneur).

Two conditions in that description matter more than the headline. The tool is specified — Gemini, not whatever assistant you prefer — and the permission attaches to one round, not the loop. A pilot scoped to particular teams and levels also means a Google interview in September 2026 may or may not include it, which is a question for your recruiter rather than for a blog post.

Meta has been testing something similar. Internal communications verified by 404 Media and reported July 30, 2025 describe AI-enabled coding interviews, with the company's stated reasoning that the format "is more representative of the developer environment that our future employees will work in, and also makes LLM-based cheating less effective" (TechRadar). That second clause is worth noting: at least one large employer treats permitting AI as a countermeasure to covert AI use, not a concession to it.

Prohibited, with disqualification stated

Amazon's internal recruiting guidance, viewed by Business Insider and reported in March 2025, instructs candidates: "To ensure a fair and transparent recruitment process, please do not use GenAI tools during your interview unless explicitly permitted." The guidance states that "failure to adhere to these guidelines may result in disqualification from the recruitment process," and Amazon also supplied interviewers with behavioral cues to watch for, including typing sounds, unnatural reading cadence, and eye movement away from the camera (IT Pro).

Goldman Sachs takes a broader position. Fortune reported on June 11, 2025 that the bank "prohibits the use of any external sources, including ChatGPT or Google search engine, during the interview process," wanting to hear applicants in their own voice (Fortune). That wording bans ordinary web search alongside AI, which is a stricter standard than most engineering employers apply.

Employer positions at a glance

These are the positions as publicly reported, with the date of the reporting. Treat each as the last confirmed state, not a guarantee of current practice.

The table is a starting point for questions, not a permission slip. Every row is reporting on a policy that the employer can revise without announcing it.

Policies reverse, sometimes within months

Anthropic is the clearest case. In May 2025 the company told applicants: "Please do not use AI assistants during the application process," reasoning that it wanted to evaluate non-AI-assisted communication. By July 2025 it had reversed that position, inviting candidates to use Claude to refine application materials, research the company, practice answers, and complete take-home assessments when instructed, while still restricting live interviews and most assessments unless explicitly permitted (Fortune).

The practical lesson is that a policy you read about in a news article is evidence about a moment, not about your interview. A ban you remember from last year may have been lifted; a permission may have been narrowed to one round. The authoritative source is the invitation you were sent.

How to tell which regime applies to your interview

Read the round name before the round description

Employers that permit or require AI generally rename the round. "AI-Assisted Coding," "code comprehension," and "AI-enabled coding interview" are labels that exist because the format changed. A round still called "coding screen," "CS fundamentals," or "algorithms" has almost certainly not been redesigned around an assistant.

Treat silence as prohibition, and ask anyway

None of the prohibiting employers above rely on candidates inferring the rule — Amazon and Goldman Sachs both put it in writing. But absence of a written ban is not permission. Ask the recruiter a narrow question in writing: may I use an AI assistant during this specific round, and if so which tools. A written answer is the only version of this that protects you, and it also settles the ambiguous middle where an employer allows documentation lookups but not generation.

The platform is not the policy

The assessment vendor sets what is technically available; the employer sets what is allowed. HackerRank, CodeSignal, CoderPad, and Codility each expose configuration that a hiring team chooses, which is why Karat's rules differ by interview module rather than applying uniformly. Where AI is enabled, the platform often records the interaction itself — HackerRank records candidate AI prompts as part of the submission, meaning the conversation is scored material rather than private scaffolding.

What AI-fluency rounds actually grade

When an employer moves AI into the interview, the thing being measured moves with it. Google's pilot names prompt engineering, output validation, and debugging. Canva names tool selection judgment, requirement decomposition, error detection in generated code, and production readiness. Meta's stated rationale points at representativeness of a real development environment.

None of those is "produced a correct answer quickly." They are all variations on whether you can direct a tool and then catch it being wrong, which is why the scoring model for AI-assisted coding interviews diverges so far from traditional algorithm scoring. Practically, that changes preparation: rehearsing a fast correct solution matters less than rehearsing how you narrate a decision, verify generated output, and reject a plausible-looking answer that does not hold.

It also creates a failure mode that did not previously exist. In a round that grades validation, accepting a confident wrong answer from an assistant is a worse outcome than not using the assistant at all.

Limitations and what remains unclear

Most of the evidence above is reporting on internal documents and company blog posts, not published policy pages that candidates can check on demand. Three gaps follow from that.

Scope is usually underspecified. Google's pilot is described by level, team, and region, but no public source lists which teams. A candidate cannot determine eligibility from outside.

Rubrics are not public. Interviewers are reported to assess AI fluency, but no employer above has published the weighting. Advice about how to optimize for those rounds is inference, including the advice in this article.

Enforcement in the prohibiting case is partly behavioral. Amazon's reported guidance trains interviewers to notice reading cadence and gaze direction, which are indicators rather than proof. That cuts both ways: it means covert use carries real risk of a judgment call going against you, and it also means candidates who are simply reading their own notes can be misread.

Where this leaves your preparation

Sort your pipeline by regime before you prepare for it. An employer that requires AI in a named round wants to see you work with a tool under time pressure, and rehearsing that workflow — including catching a wrong generated answer out loud — is legitimate preparation. An employer that prohibits AI wants unassisted work, and the preparation is the traditional kind. When you cannot tell, ask in writing and wait for the answer rather than deciding on your own reading of a news story.

Control is a desktop AI interview assistant for Windows and macOS, and it belongs in the first and third of those buckets: practicing for an AI-allowed round, and preparing before an interview rather than during a prohibited one. New accounts include a small free allowance of 5 messages and 2 minutes of voice, which is enough to see how the workflow feels; paid access starts at a $12 day pass. If your next round is one of the redesigned ones, the more useful next read is what AI-assisted coding interviews actually grade, because the rubric changed more than the rules did.