Superintelligence Rules and Regulation: What Companies Track Now
A practical watch list of policy, contract and customer signals to monitor so your company stays ready for superintelligent AI, without guessing timelines.

There is no finished rulebook for superintelligent AI that a company can simply comply with. What companies can do now is watch the channels through which obligations already arrive: laws that reach AI systems in general, the terms in vendor and customer contracts, the questions buyers and auditors ask, and the evidence the company keeps about its own systems. A short, owned list of these signals lets a business adapt when something changes, without predicting when rules will arrive or what they will require.
Key takeaways
- Treat superintelligence regulation as a set of signals to watch, not one rule to comply with.
- Track four groups: laws that already reach AI systems, contract terms, customer and buyer requirements, and your own evidence trail.
- Record each signal in one list with an owner, a review cadence and a linked action or a reason for no action.
- Keep approval boundaries and pause paths tested, because those are what you will need to show whatever the rules say.
- Build plans around controls you can operate today, not around predicted timelines.
What superintelligence regulation means for a business today
Public discussion of superintelligence covers risks that are still largely speculative. For a company, the practical question is narrower: which obligations already attach to the AI systems you run, and which new expectations could attach through the same doors. Those doors are familiar. They include data protection law, sector rules, supplier contracts, customer security reviews and the internal policies that auditors read. A more capable system would be governed through these same channels, so companies that already document purpose, data, access and oversight will find new expectations easier to absorb.
This article does not forecast when rules will change or what they will say. It describes what to watch and what evidence to keep, so that a change in any channel becomes a scheduled review rather than an emergency. For the broader risk picture, see Is Superintelligence Dangerous for Companies? A Sober Business View once you have read the sections below.
Which laws already reach your AI systems
Start with the laws that exist. Data protection law applies to any agent that reads, stores or sends personal data, whatever the underlying model. In Brazil, the LGPD sets duties for processing personal data, and the national data protection authority, ANPD, publishes guidance you can monitor directly. In the European Union, the Artificial Intelligence Act sets obligations based on the kind of AI system and its use, so whether it reaches your company depends on what the system does and where it operates. Frameworks such as the NIST AI Risk Management Framework are voluntary, but they give buyers, regulators and your own staff a shared vocabulary for risk. The OECD AI Principles offer high-level commitments that many governments refer to.
Watch for three kinds of change: a regulator publication that names your sector or your type of decision, an amendment that alters a definition your policy depends on (such as what counts as an automated decision), and a supplier notice that changes where or how a model processes data. Log each change even when it does not affect you yet, because the record shows the review took place.
- Regulator publications that mention AI, automated decisions or your sector
- Amendments to definitions used in your internal AI policy
- Supplier notices about model changes, hosting regions or data handling
- Enforcement decisions that name AI systems or automated processing
For the European question specifically, read Does the EU AI Act Apply to Your Company's AI Agents? and bring the answer to your legal reviewer rather than treating it as settled.
Which contract terms carry the obligations
Many obligations reach a company through contracts before they reach it through statute. Your AI vendor, hosting provider, integration partner and any model provider each handle part of your data, and each agreement sets the limits of what they may do with it. Check the same items in every agreement: scope of processing, subprocessors, data location, retention and deletion, breach notice, audit rights, liability and indemnity, and what happens to prompts, outputs and logs when the contract ends.
Review the terms as a set. A data processing agreement may cover personal data while a separate product policy quietly excludes logs or transcripts. Use Data Processing Agreement for AI Vendors: What to Check as the baseline, then trace the chain behind each vendor with AI Vendor Subprocessors: Who Else Handles Your Company Data?. Whenever a contract changes, update the signal list the same day so the record and the agreement do not drift apart.
What customers, buyers and auditors will ask
Customers and procurement teams often set requirements earlier and more specifically than regulators do. Watch security questionnaires, vendor onboarding forms and contract redlines for repeated questions. Each question you cannot answer with evidence today is a gap, and the list of gaps is your practical roadmap. Typical questions include:
- Which actions can the agent take without a human approving them?
- Can a named person halt the agent, and has that halt been tested?
- Can you reconstruct what the agent did, why, and with which data for a given case?
- Can personal data be removed from memory, logs and transcripts on request?
- Who is accountable inside your company when the agent makes an error?
Treat these questions as a test of your operating model. If the answer depends on a person remembering how something was set up, the gap is real even if the system works well.
Which internal evidence you must be able to produce
Regulators, insurers and auditors ask for evidence, not intentions. For an AI agent, the evidence is a set of records: a description of the role, the memory and tools the agent can reach, the permissions granted and who approved them, the points where a person must sign off, logs of actions, evaluation results over time, incident records and the procedure for pausing the agent. Each record should show a date, an owner and the version it describes.
Logs are both proof and risk. They help you show what happened, but they can contain personal data that the company must protect and delete on request. Decide retention periods and redaction rules before you need them, and test pause paths with the same care you give to any operational control. AI Agent Kill Switch: Halt Actions Without Losing Logs or Queued Work describes the pause behavior that makes this evidence usable during an incident. Quarterly permission reviews, such as those described in the access review guidance on this journal, keep the evidence current as people and tools change.
A monthly routine for tracking the signals
A signal list is only useful if someone reads it on a schedule. The following routine takes an afternoon a month for a small company and can be scaled up without changing its shape.
- Name one owner for the signal list and one backup who can run the review.
- Collect new signals from four sources: regulators, suppliers, customers, and your own incidents and reviews.
- Classify each signal as no action, watch or act, and write one sentence explaining the classification.
- Link every act item to a named control: a permission change, a contract amendment, a policy update or a new evaluation.
- Hold the review at a fixed time each month and store the decisions in the same record as the signals.
- Test one approval gate and one pause path each quarter, and record the result.
Illustrative scenario: A services company runs a scheduling agent that also reads customer messages to answer routine questions. A supplier notice says a model will be hosted in another region. The owner classifies the notice as act, the privacy lead checks the subprocessor list and the data processing agreement, and the team limits the agent to scheduling data until the review closes. The rules did not change, but the company can show the signal, the decision, the owner and the control in one record.
Be Human runs this kind of routine as part of managed operations for the agents it designs. An engagement starts with one defined role, records the memory and tools that role uses, sets approval boundaries and an observable success criterion, and then covers monitoring, recurring evaluations, permission reviews and incident response. The practical result is that the signal list and the evidence behind it exist as operating records, updated as the system runs, rather than as a document assembled before an audit.
Where to start this quarter
The first quarter does not need a complete program. It needs an accurate inventory and a working habit. Begin with these steps:
- List every AI system that touches company data, including features switched on inside tools you already use.
- For each system, note the data it reads, the actions it can take and the contract that covers it.
- Name an owner for the signal list and put a monthly review on the calendar.
- Confirm that one pause path works from start to finish, and write down what you saw.
- Mark the three questions from your customers or auditors that you cannot answer today, and assign each one to a person.
To find the first task worth handing to an agent, take the free Be Human AI diagnostic.
Frequently asked questions
Is there a finished superintelligence law that companies must follow today? No. Companies work with existing obligations, including data protection law, sector rules, supplier contracts and customer requirements, along with voluntary frameworks such as the NIST AI Risk Management Framework. The practical task is to track how those channels apply to your systems and to keep evidence that shows you reviewed them.
Should we wait for clearer rules before preparing? Waiting leaves you with no record, no owner and no tested controls when a requirement does arrive. A signal list and an evidence trail are useful under any future rule because they describe your own systems. Preparation of this kind does not require predicting what the rules will say.
What is the first document a company should create? Start with an inventory of AI systems that touch company data, with the data each reads, the actions each can take and the contract that covers it. The signal list comes next, with one owner and a monthly review. Together these two records answer most early questions from customers, insurers and auditors.
Does this apply to a small company with one or two AI agents? Yes, although the scale is smaller. A small company needs the same four things: an inventory, an owner, a signal list and a tested pause path. The routine can take an afternoon a month, and the records can live in a shared document until the operation grows.
Your next step
See where an AI agent fits in your company first.
Take the free Be Human AI diagnostic. A few questions about your work show which task to hand to an agent first, what it needs to know and where a person approves.
Be Human editorial guidance. Examples are illustrative unless explicitly attributed. Sources provide background; implementation choices depend on your workflow.


