As of January 1, 2026, the Ontario hiring landscape changed fundamentally. Under Bill 190, employers are legally mandated to disclose the use of artificial intelligence in their recruitment processes, specifically when AI is used to screen, assess, or select candidates.
For tech leaders, this creates a significant security challenge. How do you comply with transparency laws without handing a roadmap to fraudsters?
If your job posting describes your AI detection tools in too much detail, you are effectively telling Prompt Gamers exactly which filters they need to bypass. At STACK IT, we believe that compliance should never come at the cost of technical integrity.
Understanding the compliance mandate
To navigate the new regulations, you must first understand the distinction between Bill 149 vs 190. While Bill 149 laid the groundwork for AI transparency, Bill 190 introduced stricter requirements for vacancy labeling, salary disclosure, and 45-day candidate follow-up.
The AI disclosure requirement specifically triggers when an automated system plays a role in the selection or assessment of a candidate. This includes:
- Resume parsers. Tools that use AI to rank or score candidates based on keywords.
- Transcription and analysis tools. Software that transcribes interviews and provides sentiment or competency scores.
- Automated scoring tools. Any third-party software that issues a pass/fail grade based on AI analysis.
Failing to include this disclosure can lead to provincial violations and 3-year record retention audits.
However, over-sharing your methods is equally dangerous.
The security risk of over-disclosure
In our deep dive on why automated AI-detectors fail, we established that you can’t fight a bot with a bot.
When you list specific AI detection tools in your Bill 190 hiring requirements documentation, you are providing adversarial training data to dishonest candidates.
For example, if a candidate knows you use a specific vendor to check for real-time botting, they will simply use a cleaner LLM wrapper designed to avoid that vendor’s specific detection patterns. This is why STACK IT maintains a security through obscurity policy for our clients.
Drafting a protected disclosure
A compliant disclosure should be clear about the existence of AI tools but vague about the criteria and specific vendors used. This satisfies the law without inviting fraud.
The STACK IT standard disclosure for job postings:
STACK IT uses AI-enhanced tools to support initial candidate screening and interview note analysis. All assessments and hiring decisions remain human-led.
This phrasing works because it acknowledges the presence of technology (transcription via BrightHire or resume tracking via Workable) while reinforcing that the Human Insight is the final authority. It does not give the fraudster a target to hit.
What over-disclosure looks like, side by side
The argument is easier to see in the wording than in the abstract. Both versions below satisfy Bill 190. Only one of them is safe to publish.
Over-disclosed: “We use an AI-assisted screening process. Applications are parsed by our ATS, ranked against the job description using keyword and semantic matching, and candidates scoring above threshold are invited to a recorded video assessment analysed for response consistency and facial movement.”
Compliant and protected: “AI-assisted tools are used to support parts of our hiring process, including application screening. Final decisions are made by our hiring team. Contact us if you would like to know more about how your application is assessed.”
The second version tells a candidate everything the statute requires them to know. The first tells a fraudster which signals to defeat: keyword-match the description, keep answers internally consistent, and control facial movement during the recorded segment. Bill 190 asks you to disclose that artificial intelligence is used and broadly where. It does not ask you to publish your detection criteria, and nothing in the regulation requires naming specific tools.
The rule of thumb is to describe the function and never the threshold. “Screened for relevant experience” is a function. “Ranked by semantic similarity above 0.7” is a threshold, and a threshold is a target.
Maintaining forensic integrity post-disclosure
Once the disclosure is published, your primary defense moves from the job posting to the interview. Because your detection maneuvers, like these 3 physical tests to spot a deepfake candidate, are human-led actions rather than automated AI selection steps, they do not require specific vendor-level disclosure in the job ad.
We recommend a two-layered defense:
- Administrative AI: Use compliant tools for transcription and scheduling to stay organized and audit-ready.
- Human Forensics: Use the Profile Turn and Hand Pass during live calls to verify humanity in real-time.
These human-led tests are the only way to catch a high-end imposter who has already optimized their bot to pass your disclosed automated filters.
According to Deloitte’s 2026 Human Capital Trends, organizations that prioritize disinformation security, specifically in their talent pipelines, report 22% higher retention rates among top-tier talent. High-performers do not want to work alongside proxy hires, they want to be part of an elite, verified human team.
What forensic integrity means in practice
Forensic integrity sounds abstract until a hire goes wrong and someone asks how the candidate got through. What you need at that point is a record that lets you reconstruct the decision, and Bill 190 already obliges you to keep most of it for three years.
- The posting as published, including the disclosure language in force on the day it went live. Templates change, and the archived posting is what proves what you told candidates.
- Screening output as well as the outcome. If a tool ranked or scored applicants, retain the ranking alongside who advanced. A proxy interview is often visible in retrospect as an unexplained gap between screening score and interview performance.
- Interview recordings and notes, with timestamps. Proxy and deepfake detection almost always happens after the fact, and it depends on comparing a later interaction against an earlier one.
- Identity verification artefacts, stored separately from the assessment record, so verification can be re-run without reopening the hiring decision.
- Who saw what, and when. A record reconstructed from four inboxes six months later is not a record.
The three-year retention rule and fraud detection want the same artefacts for different reasons, which is convenient. A compliance archive built properly is also an investigation archive. Our guide to proxy and identity fraud covers what those signals look like in a live process, and three physical tests covers detection during the interview itself.
The case against under-disclosing
This argument has an obvious failure mode, so it is worth naming. If protecting your detection stack is the goal, the tempting move is to disclose as little as possible, or to write something so vague it says nothing. That carries its own risk, and it is the more likely of the two to actually cost you.
Enforcement under the Employment Standards Act generally starts with a candidate complaint rather than an audit. A candidate who suspects they were screened out by a tool, and finds a disclosure that told them nothing, is exactly the person who complains. Vague disclosure does not reduce that risk, it concentrates it, because you have neither the protection of saying nothing nor the credibility of saying something useful.
The line sits at usefulness to the reader. A candidate should finish your disclosure knowing that AI is involved, roughly where in the process, and that a human makes the final call. If your wording fails any of those three it is under-disclosed, regardless of how short it is. If it also tells them the scoring method it is over-disclosed. The gap between those two is wider than it looks.
A unified strategy
In the 2026 market, compliance is a baseline, but security is the differentiator. You cannot have one without the other. If your process is transparent but vulnerable, you will eventually face the high cost of a bad hire—quantified at over $50,000 per mis-hire.
At STACK IT, our success-based recruiting model integrates Bill 190 compliance directly into a forensic vetting framework. We handle the legal disclosures so you can focus on building your team with absolute certainty that every hire is both authentic and qualified.
Is your process compliant, or is it a vulnerability? Don’t let transparency become your biggest security hole. Download the Forensic AI Hiring Playbook to see our full list of detection tests and access the complete Bill 190 Ontario Hiring Compliance Checklist.