AnthROS™Lawful, on-policy — and provably so.
Every people decision on the right side of the law and your policy
AnthROS keeps every people decision — HR’s, and every line manager’s — grounded in current employment law and your own policy. It checks the decision, flags the risk, cites the rule, and leaves a record that proves it. It never invents an answer.
For HR, managers and employees. Deterministic — right by construction. A defensible record on every decision.
HR chatbots are easy to use. Handbooks are accurate but unread. Employers need both.
A general HR chatbot answers any question fluently — and will confidently quote a policy term that is actually void, or apply last year’s law. A handbook is accurate but static, unread, and silent on how statute and policy interact.
The gap between them — a reliable, current, policy-aware answer — is where the costly mistakes live: underpayment claims, grievances, and now Workplace Fairness Act exposure. AnthROS fills it: plain-language questions, a deterministic reasoning core, answers grounded in the Act and your policy.
A RAG chatbot can sound right and still be wrong. In HR, wrong is a liability.
Retrieval-augmented chatbots feel effective — fluent, fast, always ready with an answer. But retrieval still ends in generation, so they hallucinate: they state a policy term that doesn’t exist, apply a rule that was superseded, or give a confidently wrong entitlement — in the same assured tone as a correct one. In casual use that’s an annoyance. In HR it is a liability, because someone acts on the answer.
A RAG chatbot
Predicts a plausible answer from retrieved text. It can invent a policy clause, miss the statutory floor, or apply last year’s law — and a wrong answer looks exactly like a right one.
- Probabilistic
- Unverifiable
- You can't tell when it's wrong
AnthROS
Doesn’t predict the answer — derives it from the codified law and your verified policy, and cites both. Where it can’t derive an answer, it asks or flags rather than guess.
- Deterministic
- Cited to the Act and the clause
- Right by construction
The cost of a wrong HR answer is concrete: an underpayment claim, a wrongful-dismissal exposure, a discrimination breach under the Workplace Fairness Act, a decision you cannot defend. You don’t need a chatbot that is usually right — you need one that is right about the law and your policy, and shows its working.
The decision-safety layer for people decisions
The same deterministic engine serves the three people who touch employment risk — and the one who creates most of it, the line manager, is where AnthROS earns its keep.
The decision-check
Before a manager makes the call — a dismissal, a performance rating, a promotion, a hire — AnthROS checks it against employment law and your policy, flags where it’s risky, and records the reasoning. The people who create the most employment risk, kept on the right side of it at the moment they decide.
Checks the decision · flags the risk · leaves a record
The employee chatbot
Staff ask in plain language — "how much notice do I owe?", "am I entitled to childcare leave?" AnthROS answers from the Act and your policy, cites the source, and hands anything that needs judgment to HR. Self-service HR can trust, because it can’t invent an answer.
Quick, cited answers · bounded · fewer HR tickets
Guided Analysis for HR
The harder question, worked end to end — the way Guided Analysis works in MikeROS for law. AnthROS drives the intake, walks statute × policy, checks the floor, flags discretion, and produces a traceable, defensible output: a retrenchment plan, a leave determination, a WFA readiness check.
Structured reasoning · statute × policy · a defensible record
Your policy governs above the floor. The law overrides below it.
Employers almost always sit above the statutory minimum — more leave, longer notice, extra benefits — so the real answer usually lives in your policy. But the Employment Act sets a hard floor: any term less favourable to the employee than the Act is null and void (section 8). AnthROS reasons the two together. It never quotes a policy term the Act voids, and never applies the statutory minimum where a more generous policy governs. General AI tools blend the two and get this exactly wrong.
Policy governs
Where your policy is more favourable than the Act, your policy is the answer — and AnthROS cites the clause.
Section 8, hard line
The Act’s minimum is a floor no contract can drop below. AnthROS checks every policy term against it.
Statute overrides
A term less favourable than the Act is void. AnthROS flags it and applies the statutory minimum instead.
Statute, your policy, and the line where judgment takes over
Statute
The Employment Act and the Workplace Fairness Act, codified as deterministic rules — coverage, notice, leave, salary, hours, retrenchment, thresholds and time bars.
Codified once · governed
Your HR policy
Your handbook and policies, ingested and verified into rules, bounded by the statutory floor. Done per organisation, because every policy is different — not left to a model to read unchecked.
Verified · per organisation
Discretion — flagged
Whether a dismissal was fair, whether conduct was discriminatory, whether a restraint is reasonable: fact- and judgment-heavy questions are routed to a person, never answered by machine.
Human judgment · never automated
Plain language on the surface. Deterministic reasoning underneath.
AnthROS uses a language model to read the question and explain the answer — but it never lets the model decide the entitlement. The situation becomes symbolic facts and rules first; the symbolic layer derives the answer.
Neural understanding
Interprets a plainly-worded question, pulls out the facts that matter, and explains the answer in everyday language.
Reads & explains
Symbolic reasoning
Derives the entitlement from statute × policy — obligations, exceptions, the floor — deterministically, with every source pinpointed.
Decides & cites
Governed answer
Bounded to what the rules and your policy support. Where they don’t determine it, it asks, flags, or says it is out of scope.
Answers only what holds
Six steps, and a person only where judgment is required
Ask
Someone asks a question in plain language.
Clarify
It asks only the facts that matter — coverage, category, tenure, which policy applies.
Reason
It derives the answer from statute and your policy, applying the floor-and-override logic.
Answer
It returns the answer with every source pinpointed — the Act section and the policy clause.
Flag
If the question turns on discretion, or a fact is missing, it flags or asks — it does not guess.
Trace
Every answer shows its working: the rules applied, the policy used, the floor checked.
A reasoned answer a chatbot can’t give safely
“How much annual leave does she get?”
Answer: 18 days. The handbook governs because it exceeds the statutory floor — cited to the Act and to the handbook clause. Had the handbook said 5, AnthROS would flag that term as void under section 8 and apply the statutory minimum instead.
Every answer lands on one of four outcomes
Answer
The Act and your policy determine it. A cited answer is returned.
Ask
One decisive fact is missing — tenure, category, which policy applies — and can be supplied.
Flag
The question turns on discretion — fairness, discrimination, reasonableness — and goes to a person.
Out of scope
It falls outside the covered law and policy. It says so, plainly.
The general tools don’t know the new law. AnthROS does.
Singapore’s Workplace Fairness Act, completed in 2025 and taking effect around end-2027, turns decades of voluntary tripartite guidelines into binding law — a new statutory tort of discrimination, protected characteristics across the employment lifecycle, strict time bars, a mandatory grievance → mediation → adjudication route, and penalties up to S$50,000, rising to S$250,000 for repeat breaches. Every manager’s call — a hire, a rating, a dismissal — now has to be justifiable and on-policy, and an undocumented judgement is hard to defend. Ask a general model today, and it will most likely give the pre-Act answer. Fluent, and wrong.
| The question | General AI — trained on the old regime | AnthROS — the current Act |
|---|---|---|
| Can an employee sue us for discrimination? | No standalone claim — voluntary mediation only | Yes — a statutory tort of discrimination now exists |
| By when must a hiring-stage claim be raised? | Unspecified / no fixed bar | Within the Act’s strict time bar for that dispute type |
| Does the fairness law apply to us? | Guidelines — advisory, non-binding | Binding, by employer-size threshold, phased in |
| What’s the exposure for a breach? | Reputational / tripartite censure | Penalties to S$50k, rising to S$250k for repeats |
The rule-based core of employment — statute and policy together
It cannot invent an entitlement
The neural layer interprets the question and explains the answer — but it cannot produce an entitlement or a citation. An answer is returned only when the reasoner derives it from the governed statute and your verified policy. Otherwise it asks, flags, or declines. That is architecture, not a promise to behave.
The flagship demonstration for AWS’s Bedrock Automated Reasoning checks was, tellingly, a company-HR-policy hallucination — an assistant confidently inventing a 45-day leave entitlement when the real cap was ten. That is exactly the failure AnthROS is built to remove: the deep employment vertical of the same neuro-symbolic approach — a language model bounded by a symbolic reasoner — that the largest cloud provider now ships as a safeguard.
Everyone who makes a people decision — and everyone who has to defend one
Your HR systems act. AnthROS supplies the reasoned, cited answer.
AnthROS doesn’t replace your HR systems. It reasons the answer and hands it over — with the trace attached — to the systems and people who act on it.
HRIS
Workday, BambooHR, and the rest — the systems of record AnthROS answers against.
Handbook & policies
Your employee handbook and policy documents, ingested and verified into rules.
HR service desk
Employee and manager questions, answered in the channel they already use.
Statutory sources
MOM, CPF and the governing Acts — the authorities every answer is pinned to.
Bring your handbook and a few real questions
You bring
- Your employee handbook / HR policies
- 10 real HR questions your team gets asked
- Your employee categories and coverage
- Your leave and notice policies
You get back
- A structured policy-rule inventory
- The statute × policy floor-and-override map
- Reasoned, cited answers to your sample questions
- A report of where your policy dips below the statutory floor
Proven engine. Traceable answers. Named science.
Runs on the Reasonex™ engine
The same deterministic reasoning core proven on Singapore civil procedure in MikeROS™ — pointed at employment law and HR policy.
Every answer shows its working
Facts → rules → policy → answer, each source pinpointed. Verify before you rely; prove how you got there if asked.
An SUTD ARISE startup
Our Chief Scientific Officer, Prof. Ernest Chong, is on SUTD’s faculty and the originator of algebraic machine reasoning.
Built for sensitive HR data
Data governance and residency are treated as first-order requirements, not an afterthought.
Questions buyers ask
See it answer your own HR questions
Bring your handbook and a few real questions. We’ll show the statute-and-policy reasoning behind every answer, with each source cited — and where it stops and hands the question to a person.