Before You Upload That Board Paper to ChatGPT: The Legal Risks of Using Generative AI with Confidential Corporate Information
IMAGINE THIS
A director receives a 60-page board paper concerning a proposed acquisition. It contains financial projections, employee information, pending litigation, proposed purchase price, tax exposure, customer information and the company’s negotiating strategy. The board meeting is tomorrow. Instead of reading the entire document, the director uploads it to a generative Artificial Intelligence platform and asks: “Summaries this board paper and identify the ten biggest risks.” Within seconds, the AI produces an excellent summary. The director has saved several hours. But another question has quietly arisen: what exactly has the director just done with the company’s confidential information?
The legal problem with generative AI does not begin only when AI gives the wrong answer. Sometimes, the legal problem may begin the moment confidential corporate information is uploaded into the system.
1. A BOARD PAPER IS NOT JUST ANOTHER DOCUMENT
Board papers may contain acquisition proposals, financial forecasts, proposed redundancies, employee information, customer data, legal opinions, tax strategies, litigation assessments, intellectual property, pricing information and negotiations that have not yet been announced.
The convenience of generative AI can make it easy to overlook what is actually happening. When a director copies information from a board paper into an external AI platform, the question should not simply be: “Can AI summaries this?” The better question is: “Am I legally entitled to disclose this information to this system in the first place?”
2. THE COMPANIES ACT STILL FOLLOWS THE DIRECTOR INTO THE DIGITAL WORLD
The Companies Act, Cap. 212 R.E. 2023 does not expressly refer to ChatGPT, generative AI or large language models. It does not need to.
Section 185(1) requires a director, when exercising powers or performing duties, to act honestly, in good faith and in what the director believes to be the best interests of the company. That obligation does not disappear because the director is using new technology.
If a director places highly sensitive company information into an external AI system without understanding how that information will be processed, retained, accessed or protected, an important company-law question arises: was that conduct consistent with the director’s obligation to act in the company’s best interests?
Uploading a publicly available annual report is one thing. Uploading an unpublished acquisition proposal, confidential legal opinion or customer database may be something entirely different.
3. CONFIDENTIALITY IS ALSO A QUESTION OF THE DIRECTOR’S DUTY OF CARE
Section 188 of the Companies Act, requires a director to exercise the care, skill and diligence that would be exercised in the same circumstances by a reasonable person possessing the knowledge and experience reasonably expected of a person in that position, together with any special knowledge and experience possessed by the particular director.
Applied to generative AI, the question becomes: what should a reasonably careful director know before uploading confidential corporate information into an AI platform?
At minimum, the director should consider
- what information is being disclosed,
- whether the company permits use of that AI system,
- whether personal data is involved,
- whether contractual confidentiality obligations apply, and
- whether an approved enterprise environment is available.
Directors do not have to become cybersecurity engineers. But convenience is not a substitute for care. The fact that an AI platform can accept a document does not necessarily mean that a director is legally entitled to give it the document.
4. WHEN AI MEETS CORPORATE CONFIDENTIALITY AND DATA PROTECTION LAW?
The legal risks become more serious where corporate documents uploaded into generative AI systems contain personal data. Board papers, employment records, customer databases and internal reports frequently contain information relating to identifiable individuals.
Under section 3 of the Personal Data Protection Act, 2022, personal data includes information relating to an identifiable person, while processing encompasses activities such as obtaining, recording, holding, analysing and using such information. Consequently, uploading a corporate document containing personal data into a generative AI platform for analysis or summarization may constitute processing of personal data within the meaning of the Act.
Section 5 of the Personal Data Protection Act requires personal data to be processed lawfully, fairly and transparently, collected for specified and legitimate purposes, limited to what is necessary, and protected against unauthorized access or processing. Furthermore, sections 25 and 26 restrict the use and disclosure of personal data beyond its intended purpose, subject to statutory exceptions. Section 27 imposes obligations concerning data security and the responsibilities of data controllers and processors, while sections 31 and 32 regulate the transfer of personal data outside Tanzania. These provisions raise important legal questions whenever corporate documents are uploaded into AI systems operated or hosted by third-party service providers, particularly those located outside Tanzania.
However, corporate confidentiality and personal data protection are legally distinct obligations. Not every confidential corporate document contains personal data, and not all personal data constitutes a corporate secret. An unpublished acquisition strategy may be commercially confidential without involving identifiable individuals, while an employee database may attract statutory data protection obligations regardless of whether it constitutes a trade secret. Accordingly, a single AI upload may simultaneously raise questions concerning directors’ duties, contractual confidentiality, professional obligations and compliance with the Personal Data Protection Act.
The practical question for directors and corporate officers is therefore not simply whether AI can summarize a document, but whether the company is legally entitled to disclose and process the information through that particular platform. Who will access the information? Where will it be stored? Is the processing necessary and lawful? What safeguards exist against unauthorized disclosure or subsequent use? These are not merely technical questions; they are matters of legal compliance and corporate governance. The explanation that “I only uploaded the document for AI to summarise it” does not, by itself, establish compliance with the law.
5. WHAT DID THE COMPANY PROMISE OTHER PEOPLE?
Some corporate information is confidential because the company has contractually promised to keep it confidential.
Consider an acquisition where the parties sign a Non-Disclosure Agreement before exchanging financial statements, customer information, technical documents and commercially sensitive projections. If an employee or director later uploads those documents into a generative AI platform, the first contractual question is: did the NDA permit that disclosure or processing?
The same issue may arise under financing agreements, shareholders’ agreements, employment contracts, consultancy agreements and other commercial contracts containing confidentiality clauses. An AI upload may therefore have legal consequences even where the information never becomes public. The disclosure itself may require legal examination.
6. THE MOST DANGEROUS DOCUMENT MAY BE THE LEGAL OPINION
Imagine that the board receives a legal opinion concerning a potential tax exposure of TZS 10 billion. A director uploads the entire opinion into a public generative AI platform and asks: “Explain whether our lawyers are correct.”
That should immediately raise a confidentiality concern. Legal advice commonly contains highly sensitive information communicated within a professional relationship. The Advocates (Professional Conduct and Etiquette) Regulations, 2018 impose confidentiality obligations concerning client information and the client’s business and affairs.
For lawyers, the issue is even sharper. An advocate who uses AI to process a client’s instructions, evidence, transaction documents or litigation strategy must still protect the client’s confidential information. AI may assist the advocate. It does not remove the advocate’s professional obligations.
7. “BUT I REMOVED THE COMPANY’S NAME” MAY NOT BE ENOUGH
Removing a company’s name before uploading a document can reduce risk, but it may not always amount to effective anonymization. Transaction values, locations, employee identifiers, account numbers, unique contractual terms and other details may still reveal the company, transaction or individual involved.
The correct question is therefore not merely: “Did I remove the name?” It is: “Can the person, company or transaction still reasonably be identified from what remains?”
8. THE DIRECTOR SHOULD KNOW WHICH AI HE OR SHE IS USING
Not all generative AI systems operate in the same manner. Platforms may differ in contractual terms, retention arrangements, security controls, data-processing structures and enterprise protections. A company-approved enterprise AI environment is not necessarily equivalent to an employee opening a personal account on a public AI platform and uploading confidential documents.
For directors and senior management, the duty of care under section 188 makes that distinction relevant. Reasonable care may increasingly require asking: Which AI system was used, under what terms, with what information, and with whose authority?
9. THE RISK DOES NOT DISAPPEAR BECAUSE THE AI OUTPUT WAS CORRECT
Suppose the AI produces a perfect summary. No hallucination occurs, no cyberattack takes place and no information becomes publicly visible. That does not necessarily mean there was no legal problem. The concern may lie in the handling or disclosure of the information itself rather than the accuracy of the output. A director could receive an excellent answer while having used corporate information inconsistently with company policy, contractual obligations, data-protection requirements or the standard of care expected in handling sensitive material.
Nothing needs to go wrong technologically before something may already have gone wrong legally.
10. COMPANIES NEED AN AI POLICY BEFORE THEY NEED AN AI CRISIS
Simply telling employees “do not use AI” is unlikely to be a sustainable corporate strategy. A more practical approach is governance.
A corporate AI policy should identify approved AI platforms, information that must never be uploaded, when anonymisation is required, how personal data should be handled, whether legal documents may be processed, and what approval is required for sensitive information.
Particular caution should apply to board papers and minutes, unpublished financial information, merger and acquisition documents, customer and employee data, legal opinions and litigation material, passwords, tax and banking information, intellectual property, source code, pricing strategies, trade secrets and documents received under confidentiality obligations.
Companies should also train directors and employees. A policy provides little protection if the people using AI do not understand why uploading a document may itself constitute disclosure or processing of information.
11. BOARDROOM CONVENIENCE MUST NOT BECOME CORPORATE CARELESSNESS
Generative AI can be enormously useful. It can summarize reports, compare contracts, analyze information, identify inconsistencies and accelerate research. The answer is not to remove AI from the corporate environment. The answer is to use it deliberately.
Before uploading sensitive information, a director or employee should be able to answer three questions: What information am I giving away? Am I authorized to give it? Do I understand what happens to it after I press “Enter”? If those questions cannot comfortably be answered, the document probably should not yet be uploaded.
CONCLUSION: THINK BEFORE YOU PROMPT
The most significant legal risk created by generative AI may not always be the answer it gives. Sometimes, it is the information we give it.
The Companies Act, Cap. 212 R.E. 2023 continues to require directors to act honestly, in good faith and in what they believe to be the company’s best interests, and to exercise reasonable care, skill and diligence. The Personal Data Protection Act, 2022 adds another layer where corporate material contains personal information. Contractual confidentiality obligations and professional duties may add further layers.
Generative AI has made it extraordinarily easy to move information from a confidential corporate environment into an external technological system. Legally, however, easy to upload does not mean lawful to disclose. So, before a director uploads the next board paper and types, “Please summarize this for tomorrow’s meeting,” there should perhaps be one question before the prompt:
“Am I entitled to give you this document?” That may prove to be the most important question the AI never asks.
Pongolela David, Advocate
Corporate & Commercial Lawyer
