AI & automated decisions

What the AI does

• Screening search results. When a search returns short profile previews, a model from TypeSafe (or, as a fallback, an OpenAI model) estimates how well each person fits the requisition. Only the more promising profiles are retrieved in full. The rest are not added to the pipeline.
• Rule-based must-have check. Before AI scoring, full profiles are checked against the recruiter’s must-haves (job titles, skills, location, minimum tenure at a company). A candidate who fails more than half of them is placed in the pipeline as “rejected”, with the reasons shown to the recruiter.
• Scoring. An OpenAI model scores each remaining candidate from 0 to 100 against the requisition. It also gives a must-have pass or fail, strengths, gaps, a recommendation and its reasoning.
• Writing. Models draft first outreach emails and follow-ups, and classify candidate replies (for example: interested, not interested, unsubscribe).
• Talent assistant. An optional chat assistant can run these same steps when a recruiter asks it to.

What the models see

The requisition (title, description, must-haves, nice-to-haves and hiring-manager notes) and the candidate’s professional profile: name, headline, location, current company and title, tenure, work history, education, skills and summary, plus public GitHub activity and facts about their employer from our company database. We don’t collect or give the models fields such as age, gender, race or disability. Names, schools, locations and career history can still correlate with those characteristics, so scores can carry bias. Treat them as a starting point, not a verdict.

Where people are involved, and where they aren't

A person decides

• Recruiters write the requisition and the must-haves the tools apply.
• In the web app, the first email to each candidate is sent only when a recruiter clicks Approve & Send. If no draft exists yet, that click also writes the draft and sends it straight away.
• AutoTalent never makes a hiring decision, and never tells a candidate they’ve been rejected.
• Recruiters can see every score and its reasoning, and can restore any candidate marked “rejected”.

Automatic steps

• Filtering. Candidates are set to “rejected” in the recruiter’s pipeline, with no person involved, when they:
  • fail the rule-based must-have check
  • are scored as failing the must-haves when outreach is prepared
  • unsubscribe, report spam or bounce
  This is internal pipeline state. It isn’t sent to the candidate, and it can be reversed.
• Follow-ups. Up to two AI-written follow-ups are sent automatically to candidates who haven’t replied, unless the recruiter pauses them.
• Scheduling replies. If a recruiter turns on auto-scheduling, a candidate the AI classifies as interested gets a scheduling email automatically.
• Agent and API sends. Emails sent through the MCP server, CLI or API go out when the customer’s tool calls them. The talent assistant asks for confirmation before sending unless the organization turns that off.

New York City Local Law 144

NYC Local Law 144 covers “automated employment decision tools” (AEDTs). These are tools that produce a score, classification or recommendation used to substantially assist or replace discretionary hiring or promotion decisions for jobs in New York City. The duties fall on the employer or employment agency that uses the tool:

• an independent bias audit within one year before use, with a summary of the results published on its website
• notice to candidates who live in NYC, at least 10 business days before use. The notice says an AEDT will be used and what job qualifications and characteristics it assesses, and gives information about the data used and how to request an alternative process or accommodation

If you use AutoTalent scores this way for NYC roles, those duties are likely yours. The city’s rules define a “candidate” as someone who has applied for a specific position, so scoring people you are sourcing who haven’t applied may fall outside the law. Get legal advice for your situation.

What AutoTalent provides:

• this description of the tool, its inputs and outputs
• each candidate’s score, must-have result and reasoning in the app
• a CSV or JSON export of a requisition’s candidates and scores, for your records or an auditor

EU AI Act

The EU AI Act lists AI systems used to recruit or select people as high-risk (Annex III, point 4), including systems that filter applications and evaluate candidates. High-risk systems carry duties for:

• the provider: risk management, data governance, technical documentation, logging, human oversight design, accuracy and a conformity assessment
• the deployer, meaning the employer: using the system as instructed with human oversight, monitoring it, keeping logs, and informing the people it’s used on

These duties apply on the dates set in the Act, and the EU has proposed moving some of them. Separately, the GDPR (Article 22) limits decisions based solely on automated processing that significantly affect people.

Colorado AI Act

Colorado’s AI Act (SB 24-205, as amended) covers “high-risk” AI systems that make, or are a substantial factor in making, consequential decisions, including employment decisions.

• Deployers, meaning employers, need a risk-management policy and impact assessments. They must tell people when such a system is used, explain adverse decisions, and offer a chance to correct data and to appeal to a human.
• Developers must give deployers documentation and publish a statement about their systems.

The law’s start date has already been pushed back once, from February 1, 2026 to June 30, 2026, and lawmakers have kept revising it, so check its current status. If you hire in Colorado with AutoTalent, the sections above describe the system for your impact assessment and notices, and we can answer questions.

Other laws to know about

• Illinois: from January 1, 2026, the Illinois Human Rights Act bars employers from using AI in ways that discriminate in hiring (including using zip codes as a proxy), and requires them to tell people when AI is used.
• California: Civil Rights Council regulations, in effect since October 1, 2025, apply anti-discrimination law to automated-decision systems used in employment. They also require employers to keep related records for four years.
• Anti-discrimination law generally (for example, Title VII and the ADEA in the US) applies to hiring decisions whether or not a tool helped make them.

If you're a candidate

You can ask what data we hold about you, have it corrected or deleted, or ask a person to review how you were assessed. Email support@autotalent.ai or read the Candidate privacy notice.