May 6, 2026: AI's Real Costs Are Now on the Table — In Court, in Boardrooms, and in Bank Compliance Rooms

The past 48 hours have been unusually rich for anyone paying attention to where AI is actually going. Not the headline hype, but the operational reality. Compute budgets revealed under oath. Major employers cutting headcount explicitly because of AI. Government security reviews of unreleased models. An AML agent heading into bank compliance rooms. Two threads running in parallel: massive capital going in, and serious workforce and governance restructuring coming out.

Here's what happened.

OpenAI Just Put Its $50 Billion Compute Bill on the Public Record

OpenAI president Greg Brockman took the witness stand again Tuesday in the ongoing trial with Elon Musk, and the most significant number he delivered wasn't about damages or ownership, it was about spending. Brockman told the court that OpenAI expects to spend $50 billion on computing power in 2026 alone. For context, the company's entire compute budget in 2017 was roughly $30 million.

That 1,600x increase in nine years tells you something important about the economics of frontier AI that no press release ever quite captures. This isn't a software business with typical margins. It's a compute-intensive infrastructure play that burns capital at a scale previously associated with semiconductor fabs and satellite networks. Bloomberg and Reuters both confirmed the figures. The trial itself — Musk is seeking over $100 billion in damages — is now functioning as an involuntary transparency event for OpenAI's finances.

For enterprise teams evaluating AI vendors: the compute cost picture matters for long-term pricing stability and model availability. Companies spending at this scale need revenue to match, and that eventually flows through to enterprise contracts.

Coinbase Cut 700 Jobs and PayPal Followed Within Hours

Tuesday was a hard day for fintech headcount. Coinbase announced it would lay off approximately 700 employees — about 14% of its roughly 5,000-person workforce — with CEO Brian Armstrong framing the move explicitly around AI. His language was notable: he described the goal as building "AI-native talent who can manage fleets of agents" and experimenting with "one-person teams" where a single employee combines engineer, designer, and product manager roles. He used the phrase "rebuilding Coinbase as an intelligence, with humans around the edge aligning it" — one of the more explicit statements yet from a major tech CEO about treating the company itself as an AI system.

Within hours, PayPal's new CEO signaled plans to cut roughly 20% of the workforce over the next two to three years, also citing AI-driven efficiency. Block, which cut 50% of its staff in February under Jack Dorsey citing secular AI change, has since seen its stock rise about 38%. Coinbase fell 2.6% on the news; PayPal dropped as much as 12%.

The pattern across crypto and fintech is now consistent enough to stop calling it coincidental. Block, Gemini, Crypto.com, Coinbase, PayPal — all within roughly the same window, all citing AI. The honest read: some of these cuts are legitimately AI-driven productivity gains, and some are using AI as cover for cuts that would have happened anyway. The question worth asking before your next headcount planning cycle is which category your own efficiency story falls into — and whether your metrics can actually support that answer.

FIS and Anthropic Are Deploying an AI Agent Into Bank Compliance Rooms

On May 4, FIS — the financial technology company that processes transactions for roughly 12% of the global economy — announced a partnership with Anthropic to build a Financial Crimes AI Agent targeting anti-money laundering operations. The agent is designed to compress AML investigations from hours to minutes by automatically assembling evidence across a bank's core systems, transaction history, and customer activity.

BMO and Amalgamated Bank are currently in development with the tool, and broader availability to FIS clients is planned for the second half of 2026. Anthropic's Applied AI team embedded forward-deployed engineers directly inside FIS to co-design the agent and build the evaluation frameworks — a model that's increasingly common when AI moves into regulated, high-stakes environments where getting the architecture wrong isn't just a product problem, it's a legal one.

The governance layer here is worth noting: every conclusion the agent reaches links back to source data, and every decision stays with the human investigator. That's not a feature — it's a regulatory necessity. If your team is evaluating AI agents for compliance, audit, or risk functions, the FIS/Anthropic architecture is a reasonable reference model for what "human-in-the-loop in a regulated setting" actually looks like in practice.

Microsoft, Google, and xAI Opened Their Pre-Release Models to Government Security Teams

The Department of Commerce's Center for AI Standards and Innovation announced on Tuesday that Microsoft, Google, and xAI have signed agreements to give the federal government early access to their AI models — before public release — for national security testing. The trigger was Anthropic's Mythos model, which pushed cybersecurity concerns about advanced AI capabilities to a level that prompted White House consideration of a formal pre-launch review process for frontier models.

The deal allows CAISI to probe the models with reduced or even disabled safeguards in order to assess national security-related capabilities and risks. This follows a separate announcement from May 1 in which the Pentagon formalized AI deployment deals with seven companies — Google, Microsoft, Amazon Web Services, Nvidia, OpenAI, Reflection AI, and SpaceX — for use on classified networks. Anthropic is conspicuously absent from both lists after refusing to allow its models to be used for mass surveillance or autonomous weapons.

The implication for enterprise AI procurement teams is worth tracking: the government review process that's forming here will likely become a de facto evaluation framework that influences how regulated industries assess frontier models. What CAISI finds in these reviews may eventually surface as procurement guidance or compliance requirements.

McKinsey Is Using AI Agents to Staff Its Own Client Teams

McKinsey, which has grown to nearly 40,000 employees, announced plans to deploy AI agents to assist in matching consultants to client assignments — a role previously handled entirely by professional development employees who've done this work behind the scenes for decades. The rollout is starting in Latin America and North America, with global expansion planned by end of summer.

The goal, per Chief People Officer Wendy Miller, is for professional development employees to spend less time on administrative matching work and more time on counseling and coaching — a genuine shift in the role rather than an elimination of it. McKinsey is also one of the firms that deployed 25,000 internal AI agents across its operations and previously cut around 200 internal tech and support employees after automating non-client-facing work.

The staffing question this raises for other professional services firms is straightforward: if the firm that advises companies on organizational design is using AI to manage its own talent deployment, the conversation about AI's role in knowledge work staffing is no longer theoretical. The timeline matters too — a global rollout by end of summer 2026 is a tight implementation window for a system that affects every client engagement the firm runs.

Jamie Dimon Put a Number on What AI Is Worth

JPMorgan Chase CEO Jamie Dimon made his position explicit on Tuesday: AI will ultimately be worth more than the $1 trillion the industry is on track to invest in it. Dimon has been one of the more consistent voices arguing that AI's value to financial services is underestimated rather than overhyped — a view shaped by JPMorgan's own internal deployment across trading, risk, and operations.

That's a notable framing given the current environment. With Big Tech's combined AI capital expenditure for 2026 running toward $725 billion across just four major hyperscalers, and OpenAI's compute bill hitting $50 billion on its own, the question of whether the returns will justify the investment is the defining business question of this cycle. Dimon's view is essentially that they will — and that the companies positioned to capture that value are the ones building now, not waiting for the cost curve to fall.

Enterprise Leaders Say AI Is Delivering Value — and AI Accountability Is Now the Biggest Blocker

The Jitterbit 2026 AI Automation Benchmark Report, released Tuesday, offered a useful data point for anyone tracking enterprise AI maturity. According to the report, 78% of AI projects are now delivering real business value — effectively declaring the agentic pilot era over. The bottleneck has shifted. The biggest obstacle to scaling is no longer the CFO; it's the CISO.

Forty-seven percent of respondents identified "AI accountability" — encompassing security, auditability, and guardrails — as the single most important factor when evaluating new tools. Agent sprawl and what the report calls "agent contamination" are named as real threats to enterprise deployments. A separate Mayfield survey of 266 CIOs, CTOs, and CISOs found that over 72% of enterprises are either in production with or actively piloting agentic AI, with security and risk controls cited as the most frequently noted obstacle to full-scale deployment.

The practical read: if your agentic AI initiatives are stalling, the delay is probably governance, not capability. Teams that build the audit trail and accountability layer into their agent architecture from the start — not as a retrofit — are moving faster to full deployment. The FIS/Anthropic model mentioned earlier is one example of how that gets done in practice. Getting your CISO aligned early, with a clear data lineage and override model, is now the unlock — not getting a bigger model or a faster pipeline.

If you want to stay ahead at the intersection of AI, automation, and human performance — where technology meets psychology, processes, and real workplace behavior — subscribe to Agenticism. We cut through the hype to deliver practical insights for leaders focused on making people, processes, and technology work better together.