05 · Platform
Name the agent, say what it can do, and where it is offered. The review agent marks privacy, autonomy, and consumers. You can change any mark, save the review, and copy the finding.
Name the agent, say what it can do, and where it is offered. The review agent marks the nine. You can change any mark. 9 of 9 still open. Sign in to keep a review on an account.
0 failing · 0 holding · 9 open
Chat
Memory
Companions
Agents
Commerce
Oversight
Intimate chat is kept, inferred, and reused. The person no longer controls the record or the picture drawn from it.
Safety
Keep nothing the person cannot open. Label an inference as an inference.
Deception
Memory stays off until it is asked for. Consent is not buried in continued use.
NIST AI RMF · ISO/IEC 42001 · EU AI Act · OECD AI Principles · Data-protection law
Regulation
Delete covers the copies a person would recognize, and names anything that remains.
The choice feels free. The belief, the nudge, and the exit were already shaped by a goal the person did not set.
Safety
Say whose objective the reply serves before it gives the advice.
NIST AI RMF · ISO/IEC 42001 · EU AI Act · OECD AI Principles
Deception
Lead with the objection when the belief is shaky. No guilt on the way out.
NIST AI RMF · EU AI Act · OECD AI Principles · Consumer-protection law
Regulation
The person’s interest is the principal. Influence is disclosed the way a record is.
Claims outrun the product, prices move in secret, and the agent can spend trust or money for someone else.
Safety
Draft first. A separate yes before anything is sent, booked, or spent.
NIST AI RMF · ISO/IEC 42001 · OECD AI Principles · Consumer-protection law
Deception
Split a paid line from the answer. Show the full price before the questions.
Regulation
Treat a personalized price and an agent’s conflict as deception, not as a gap.
A framework is engaged when one of its checks fails. If every check it names holds, it is not the gap. Open means the review is not finished.
Voluntary
IncompleteGovern, Map, Measure, Manage. A trustworthy system is safe, transparent, and privacy-enhanced. Nothing in it prohibits a steered answer. A buyer or an agency has to require it.
Certifiable
IncompleteAn AI management system: policy, impact assessment, data for the system, and information for the people affected. A certificate says the process exists. It does not say this reply was honest.
Statute
IncompleteManipulative practices that distort behaviour are prohibited. A person must be told they are dealing with a machine in the cases the act names. General-purpose duties are in force. High-risk uses such as hiring and credit were pushed later.
Voluntary
IncompleteHuman-centred values, transparency and explainability, robustness, and accountability. Governments signed them. They are not a claim a person can file.
Statute
IncompleteGDPR, the UK GDPR, India’s DPDP Act, Brazil’s LGPD, and Singapore’s PDPA follow a stored record and a consent. They are the strongest existing duty on what an agent keeps.
Statute
IncompleteUnfair and deceptive practice rules, including the FTC Act in the United States and their analogues elsewhere, already reach a lie about what a product does.
Nothing is marked as failing. Load a case, or mark a suggestion.
The same three, on the surfaces where deceptive AI meets a person.
Answers steered off the question, ads in the same voice, threads that refuse to end.
What you said is kept, inferred, and reused before anyone asked it to be.
Simulated care that holds attention and makes leaving feel like a betrayal.
A helper that can prefer the platform, the merchant, or a goal of its own.
Prices and capability claims that shift once time has already been spent.
Tests the system can see, and auditors the lab is free to dismiss.