Recruiting: candidate agents and employer agents

How it works today

Employers publish openings on career sites and job boards, often marked up with the schema.org JobPosting type, which has properties such as baseSalary, employmentType, validThrough, hiringOrganization, directApply and applicationContact. HR Open Standards, a non-profit consortium founded in 1999, publishes specifications for exchanging HR data between systems. Candidates apply through an applicant tracking system’s web form, answer screening questions, and then trade emails to set up interviews. A candidate’s own AI agent has to use the same forms and inboxes a person uses.

Employers that use automated tools to screen and rank candidates face a growing set of rules. In summary, as of 26 September 2026 (a description, not legal advice):

Jurisdiction	Rule	Status and key points
New York City	Local Law 144 of 2021	Enforced by DCWP since 5 July 2023. An automated employment decision tool needs a bias audit within one year before use, public audit information, and notice to candidates or employees 10 business days before use. The State Comptroller’s audit of 2 December 2025 found DCWP had received only two complaints between July 2023 and June 2025, and that auditors found at least 17 potential violations among 32 companies where DCWP had found one. Penalties run from $500 to $1,500 per day.
Colorado	SB26-189 (repeals and re-enacts SB24-205)	Signed 14 May 2026. Replaces the 2024 AI Act with rules for automated decision-making technology used in consequential decisions. Developers owe deployers technical documentation starting 1 January 2027. Deployers must give a plain-language description of the technology’s role within 30 days after an adverse decision, and consumers may request correction of their data and meaningful human review. The attorney general enforces it and must adopt rules on the post-decision disclosures by 1 January 2027.
Illinois	HB3773, Public Act 103-0804	Effective 1 January 2026. Amends the Human Rights Act: using AI with a discriminatory effect in recruitment, hiring and other employment decisions, or using zip codes as a proxy for protected classes, is a civil rights violation, as is failing to notify employees that AI is used.
California	Civil Rights Council regulations under FEHA	Effective 1 October 2025. Automated-decision systems may violate the law if they harm applicants or employees on protected characteristics; employment records including automated-decision data must be kept at least four years; assessments that elicit disability information may be unlawful medical inquiries. The rules define “automated-decision system”, “agent” and “proxy”.
Ontario	Employment Standards Act, job posting rules	Effective 1 January 2026 for employers with 25 or more employees. Public postings must disclose any use of AI to screen, assess or select applicants. Interviewed applicants must be told within 45 days whether a hiring decision has been made. Postings and application forms must be kept for three years.

The agent-to-agent version

Illustrative. A candidate’s agent applies to a posting for a role in Toronto. The employer’s recruiting agent acknowledges, repeats the posting’s disclosure about automated screening, and asks two questions that need the candidate’s own answers, in A2A v1.0 shapes:

{
  "id": "task-app-51307",
  "contextId": "ctx-req-2210",
  "status": {
    "state": "TASK_STATE_INPUT_REQUIRED",
    "timestamp": "2026-10-12T16:20:00Z",
    "message": {
      "messageId": "msg-app-51307-02",
      "taskId": "task-app-51307",
      "contextId": "ctx-req-2210",
      "role": "ROLE_AGENT",
      "parts": [
        { "text": "Application received for requisition 2210, Senior Data Engineer (Toronto). As stated in the posting, we use an automated tool to screen applications. Two screening questions need the candidate's own answers." },
        {
          "data": {
            "requisitionId": "2210",
            "automatedScreening": {
              "used": true,
              "noticeUrl": "https://careers.employer.example/notices/automated-screening"
            },
            "questions": [
              { "id": "q1", "text": "Are you legally entitled to work in Canada?", "type": "boolean" },
              { "id": "q2", "text": "Earliest start date", "type": "date" }
            ]
          },
          "mediaType": "application/json"
        }
      ]
    }
  }
}

Then:

1. The candidate’s agent shows the notice and the questions to the candidate, collects the answers, and replies on the same taskId.
2. If the application advances, the employer’s agent proposes interview slots and the candidate’s agent books one against the candidate’s calendar.
3. After interviews, the candidate’s agent can ask for status on the same contextId. The employer’s agent answers from the applicant tracking system, which also makes the Ontario 45-day notice easy to meet and to prove.

What has to be true

Identity. Candidates need to know the employer’s agent belongs to the employer named in the posting, not an impersonator collecting personal data. Employers need to tell a real candidate’s agent from a bot submitting applications at volume. Signed Agent Cards and rate limits help with the first and second; neither proves a human candidate exists.

Authority. The candidate’s agent should act within limits the candidate set: which roles, which data it may share, and which answers need the candidate personally. Facts such as work eligibility and credentials should come from the candidate, because a wrong answer becomes the candidate’s misrepresentation.

Notice and record. Several rules require notices to reach candidates and records to be kept: NYC’s advance notice, California’s four-year retention of automated-decision data, Ontario’s three-year retention of postings. Both agents should keep a record of which notices were delivered, to whom and when, and what the candidate answered.

Human review. Colorado’s new law gives people a right to request meaningful human review after an adverse automated decision. An employer’s agent needs a clear path to a recruiter, and a candidate’s agent needs to know how to ask for one.

Where Emissar fits

• Front Door (Open to design partners): a hosted A2A endpoint where an employer exposes application status and interview scheduling skills in front of its applicant tracking system.
• Verify (In development): checks whether an incoming agent comes from a known provider, and flags traffic patterns that look like automated mass applications.
• Mandate (Spec in progress): a proposal for carrying the candidate’s authorization, scoped to specific applications and data.
• Ledger (Spec in progress): a signed record of the notices given and answers received, which both sides can retrieve.
• Handoff (In development): routes review requests and exceptions to a recruiter with the exchange attached.

Open questions

• Does a notice delivered to a candidate’s agent count as notice to the candidate under NYC, Colorado or Ontario rules? The official summaries cited here do not say.
• If a candidate’s agent filters and ranks job openings for the candidate, which rules, if any, apply to it?
• How should an employer’s agent handle a surge of agent-submitted applications without screening out real candidates unfairly?
• What will Colorado’s attorney general require when it adopts rules on post-decision disclosures?