Karsten Wenzlaff, Advisor
August 26th, 2025
March 2 2026 | Feature | AI Governance And Public Safety

AI Image: Conceptual illustration of AI ethics, defense contracts, and public accountability
On February 26 2026, Anthropic said it wouldn't allow two Pentagon use cases in its contracts: mass domestic surveillance and fully autonomous weapons. Two days later, OpenAI said it reached a Pentagon deal for classified AI deployments and argued its safeguards could still hold inside the agreement. The split exposes the real issue now facing AI vendors, governments, and regulated buyers. That is should ethics limits block deals or be enforced from inside it?
Anthropic wasn't on the outside looking in. It said Claude was extensively deployed across the Department of War and other national security agencies and that the company had already put models into classified U.S. government networks. Anthropic also said two use cases, mass surveillance (including of the public), and fully autonomous weapons, we're never included in its contracts. Those cases are the fault lines in the dispute with Trump's Pentagon and US government.
Anthropic framed the issue as a narrow refusal, not a broad rejection of national security work. Dario Amodei said Anthropic supported “98% or 99%” of Pentagon use cases, but the company would not give up the remaining two limits. In its own words, Anthropic said, “we cannot in good conscience accede to their request.” Reuters reported that stance put a defense contract worth up to $200 million at risk.
The dispute escalated when President Trump directed the government to stop work with Anthropic and ordered a six month phase out of Anthropic technology across government. The Pentagon also moved to treat Anthropic as a supply chain risk. That raised the cost of holding hard limits when the customer is the government.
OpenAI took a different approach. In its own statement titled, 'Our agreement with the department of war', the company said its agreement had “more guardrails than any previous agreement for classified AI deployments, including Anthropic’s.” OpenAI also said, “We were—and remain—unwilling to remove key technical safeguards.” That is the core tension in this story.
Anthropic refused the terms it saw as too open. OpenAI signed and argued that stronger controls inside the agreement could still hold the line.
The details matter. OpenAI said the Department of War may use its system for all lawful purposes, but the same published terms say the system cannot independently direct autonomous weapons where law or policy requires human control, cannot make high stakes automated decisions that require human approval, and cannot be used for unconstrained monitoring of U.S. persons’ private information. OpenAI also said it keeps its safety stack in place, limits deployment to cloud infrastructure, keeps cleared personnel in the loop, and could terminate the contract if the government violates the terms.
The two companies are close on headline principles but not identical in practice. Anthropic draws two hard exclusions and refuses to move them. OpenAI accepts a broader legal framework, then relies on contract language, technical controls, and operational oversight to make similar limits enforceable inside the deal. That is a narrower and more conditional ethics model than refusing the agreement outright.
North of the border, the same trust question is already under pressure. After the Tumbler Ridge school shooting, Ottawa’s focus was whether AI evidence and escalation controls hold when governments ask for proof that safeguards, response rules, and accountability processes actually work under pressure. That's different than the Pentagon story that's testing how AI companies react and accept to full on defense contracts.
For fintechs, banks, and infrastructure providers, the lesson is practical. Buyers in regulated markets now ask more than whether a model performs well. They ask who can audit it, who can override it, what records exist, how long they are kept, and whether a vendor can prove that a stated limit still holds when a regulator, a court, or a government agency pushes hard.
On one hand, Anthropic's founders' moral compass highlights one method by maintaining hard limits even at the risk of losing a major federal contract. On the other hand, OpenAI's deal with the department of war is betting that contract terms, technical safeguards, and human oversight can preserve similar limits inside the agreement. Washington shows how quickly the state can force the issue. Ottawa shows what happens when public safety puts the same claims under direct scrutiny. In the next phase of AI adoption, trust rests less on capability claims and more on what a company can prove when the pressure rises.
The National Crowdfunding & Fintech Association (NCFA Canada) is a financial innovation ecosystem that provides education, market intelligence, industry stewardship, networking and funding opportunities and services to thousands of community members and works closely with industry, government, partners and affiliates to create a vibrant and innovative fintech and funding industry in Canada. Decentralized and distributed, NCFA is engaged with global stakeholders and helps incubate projects and investment in fintech, alternative finance, crowdfunding, peer-to-peer finance, payments, digital assets and tokens, artificial intelligence, blockchain, cryptocurrency, regtech, and insurtech sectors. Join Canada's Fintech & Funding Community today FREE! Or become a contributing member and get perks. For more information, please visit: www.ncfacanada.org
![]() | ![]() | ![]() |
|---|---|---|
![]() | ![]() | ![]() |
February 17 2026 | Market Insight | Artificial Intelligence Governance

Image: Freepik
On February 17 2026, a New York federal court ruled in United States v. Heppner that written exchanges between a defendant and Anthropic’s Claude are not protected by attorney client privilege or the work product doctrine.
The case centred on about 31 documents that recorded the defendant’s interactions with Claude. Investigators collected the files during a search after the defendant received a grand jury subpoena and understood he was under investigation. He used Claude to generate reports that outlined possible defence strategy and potential legal and factual arguments. He did this on his own rather than at the direction of his lawyer.
The court applied standard privilege rules and found the documents fell outside protection. Judge Jed S. Rakoff explained that discussions between non-lawyers were not privileged and that privilege depended on a confidential relationship with licensed counsel. The court also found the Claude conversations were not confidential because they involved a third party platform and the platform’s privacy policy alerted users that data could be collected and disclosed.
HSF Kramer’s analysis of the ruling explains how client side use of public generative AI tools could weaken privilege claims. The defendant argued that he created the Claude reports to prepare for conversations with his lawyer and later shared them with counsel. The court rejected that argument. Because the defendant used the tool independently and Claude disclaimed providing legal advice, the documents did not become privileged simply because counsel received them later.
The court reached the same result under the work product doctrine. That doctrine protects materials prepared by or at a lawyer’s direction to safeguard legal strategy. Since the defendant created the documents himself, the court found they did not reflect counsel’s mental impressions or strategy.
The defendant’s approach reflected a workflow many teams now follow. People often use generative AI to organize facts, test arguments, or prepare notes before speaking with a lawyer. In practice, AI can feel like a private drafting space. The Heppner ruling shows that this assumption creates risk because privilege depends on lawyer involvement and confidentiality at the time material is created, not later.
This decision shows how everyday AI use can create governance risk. Fintech teams often rely on generative AI to draft internal notes, explore regulatory interpretations, test legal arguments, and shape product thinking.
When teams create prompts and outputs outside privileged workflows, those materials may be treated like ordinary internal documents and may surface in litigation, investigations, or regulatory reviews.
The Heppner outcome also challenges a common assumption that AI drafts become privileged once shared with a lawyer. Timing and intent matter. If teams generate strategy or legal framing before counsel becomes involved, privilege arguments may already weaken. Enterprise AI environments may offer stronger confidentiality controls, but the decision shows privilege still depends on lawyer involvement and clear intent to obtain legal advice.
If generative AI captures early legal and strategic thinking, should fintech governance treat AI prompts with the same care as internal legal memos and executive communications?
Privilege depends on how and when material is created, not what you do with it later.
The National Crowdfunding & Fintech Association (NCFA Canada) is a financial innovation ecosystem that provides education, market intelligence, industry stewardship, networking and funding opportunities and services to thousands of community members and works closely with industry, government, partners and affiliates to create a vibrant and innovative fintech and funding industry in Canada. Decentralized and distributed, NCFA is engaged with global stakeholders and helps incubate projects and investment in fintech, alternative finance, crowdfunding, peer-to-peer finance, payments, digital assets and tokens, artificial intelligence, blockchain, cryptocurrency, regtech, and insurtech sectors. Join Canada's Fintech & Funding Community today FREE! Or become a contributing member and get perks. For more information, please visit: www.ncfacanada.org
![]() | ![]() | ![]() |
|---|---|---|
![]() | ![]() | ![]() |
February 23, 2026 | NCFA Market Activity | Artificial Intelligence And Data

On February 21, 2026, B.C.’s premier says police are pursuing preservation orders for potential evidence held by digital services companies, including social media platforms and AI companies, after media reports tied OpenAI to the case. This isn't a new fintech policy just yet, but it puts evidence readiness into the open. If a partner asks what you can preserve and produce under lawful process, you can’t appear like a 'deer in headlights'. You need a working path.
BC Government, Office of the Premier, Media Relations:
“Reports that allege OpenAI had related intelligence before the shootings in Tumbler Ridge took place are profoundly disturbing for the victims’ families and all British Columbians. We will use all powers of government to ensure that police have the tools they need to investigate every aspect of this horrific tragedy.”
RCMP report said that the active shooter call came in at about 1:20 p.m. on February 10 and the public alert ended at 5:45 p.m. Six victims were found deceased inside the school, two victims were airlifted with serious or life threatening injuries, a third victim died during transport, and about 25 others were assessed for non life threatening injuries. RCMP also reports two additional victims were found deceased at a connected residence.
On February 13, RCMP reports more than 80 interviews took place with students, educators, and first responders, while forensic teams continued processing two scenes and a vehicle connected to the suspect. The investigative update also reported digital evidence work including witness images and videos, CCTV, and body worn camera video, and it says RCMP launched an online portal to collect more evidence, including phone footage captured inside the school.
In a BBC report, OpenAI spokesperson said that they didn't alert authorities about the account because its usage didn't meet its threshold of credible or imminent plan for serious physical harm to others...
"In June 2025, we proactively identified an account associated with this individual [Jesse Van Rootselaar] via our abuse detection and enforcement efforts, which include automated tools and human investigations to identify misuses of our models in furtherance of violent activities."
Evan Solomon, Minister of Artificial Intelligence and Digital Innovation per Globe and Mail reporting:
"Canadians expect online platforms, including OpenAI, to have robust safety protocols and escalation practices in place to protect online safety and ensure law enforcement are warned about potential violence,"
This story turns AI governance into an evidence workflow. If your product runs AI in onboarding, fraud triage, collections, or support, partners need clear answers on what the system records, what it retains, who can access it, and how the team preserves and produces records under lawful process without breaking integrity. A stakeholder may accept model error but they won’t accept uncertainty about records, access, and escalation ownership when scrutiny rises.
Financial technology companies should define escalation triggers in plain language. Name an accountable owner and a backup. Restrict log access. Set retention windows you can defend. Test a short runbook for preservation and production so the team can execute it without debate when the question lands, an be sure to provide suitable ongoing training.
If preservation orders become a more common tool when AI enters an incident, does Canada need a shared baseline for AI logging, retention, and escalation handoffs so every regulated buyer does not rebuild the same checklist from scratch?
Reuters reported that OpenAI reps have been summoned to Ottawa urgently to discuss AI safety.
Update per CTV article: Following the meeting in Ottawa with OpenAI safety leaders, federal officials said they were “deeply disturbed” and ultimately “disappointed” that warning signals tied to the Tumbler Ridge shooter were not escalated to police earlier, with AI Minister Evan Solomon stressing Canadians expect platforms to maintain “robust safety protocols and escalation practices.” OpenAI is cooperating with the RCMP but produced no immediate new safeguards, leaving ministers to seek concrete follow-up proposals.
The National Crowdfunding & Fintech Association (NCFA Canada) is a financial innovation ecosystem that provides education, market intelligence, industry stewardship, networking and funding opportunities and services to thousands of community members and works closely with industry, government, partners and affiliates to create a vibrant and innovative fintech and funding industry in Canada. Decentralized and distributed, NCFA is engaged with global stakeholders and helps incubate projects and investment in fintech, alternative finance, crowdfunding, peer-to-peer finance, payments, digital assets and tokens, artificial intelligence, blockchain, cryptocurrency, regtech, and insurtech sectors. Join Canada's Fintech & Funding Community today FREE! Or become a contributing member and get perks. For more information, please visit: www.ncfacanada.org
![]() | ![]() | ![]() |
|---|---|---|
![]() | ![]() | ![]() |