AI This Week: The Government Steps In
The biggest AI story this week wasn’t a new model — it was Washington deciding it needs to review models before you get them. Alongside that, API prices fell off a cliff and new data confirmed that most small businesses have already picked a side.
The U.S. Government Will Now Test AI Models Before They Launch
On May 5, the Center for AI Standards and Innovation (CAISI) — a division of the U.S. Department of Commerce — announced pre-deployment testing agreements with Google DeepMind, Microsoft, and Elon Musk’s xAI. These deals build on existing agreements with OpenAI and Anthropic, meaning every major frontier AI lab now has a government review channel before new models go public.
The evaluations will assess “frontier AI capabilities” and security risks, including a model’s ability to identify or exploit cybersecurity vulnerabilities. According to Fortune, the policy shift was accelerated by concerns over Anthropic’s next-generation “Mythos” model — a notable pivot for an administration that entered 2025 explicitly favoring minimal AI regulation.
The White House is also weighing an executive order to formally establish an AI oversight working group, with tech executives and government officials in the room together.
What this means for your business: The immediate practical effect is that major model releases may take longer. The longer-term effect matters more: the AI tools embedded in your CRM, your recruiting software, and your customer service stack are about to carry more documented accountability. When your vendors say their AI is “safe,” they’ll soon be expected to prove it to a federal agency. That’s a floor rising under the entire industry.
DeepSeek Just Cut API Prices by 75% — and It’s Not Alone
Chinese AI lab DeepSeek announced a 75% promotional discount on its V4-Pro model through May 31, and simultaneously cut cache-hit pricing across its entire API to one-tenth of previous levels. According to The Next Web, at standard pricing DeepSeek-V4-Pro already undercuts OpenAI’s GPT-5.5, Anthropic’s Claude Opus 4.7, and Google’s Gemini 3.1 Pro on a per-token basis.
This is part of a broader collapse in AI API costs. A May 2026 pricing comparison shows that costs across major providers have fallen 40–70% in the past year. Meanwhile, consumer subscription pricing has converged: ChatGPT, Claude, Google AI Pro, and Perplexity all sit at $20/month as of this week.
The competitive pressure is real. DeepSeek’s model prices its V4-Pro at 97% below OpenAI’s GPT-5.5 at standard rates, according to the South China Morning Post. Developers and agencies building custom AI workflows are already switching.
What this means for your business: If you’re paying an agency or developer to build AI-powered features, their underlying material costs just dropped dramatically. Ask directly whether those savings are being reflected in your invoice. If you’re managing your own AI API usage, a cost audit this week is worth 30 minutes — you may be able to cut spend significantly without changing a single thing about your workflow.
60% of Small Businesses Now Use AI — The Undecided Window Is Closing
New data from the U.S. Chamber of Commerce confirms that 60% of small businesses now use AI, more than double the share from 2023. The average SMB deploys a median of five AI tools, concentrated in marketing, content creation, and workflow automation.
The SBE Council’s 2026 Small Business Tech Use Survey adds detail: 82% of small business employers have actively invested in AI tools, and 93% of current AI users plan to continue or increase that investment next year. Marketing is the top use case, delivering measurable ROI in time savings and customer reach.
What changed between 2023 and now? Lower prices, better defaults inside tools people already use, and enough case studies that the ROI math is no longer theoretical.
What this means for your business: If you’re still in the 40% not using AI, the gap between you and your competitors is compounding every quarter. The entry bar is low — a $20/month subscription applied to one specific, repetitive task is a real starting point. The businesses reporting the strongest returns aren’t using ten AI tools; they’re using two or three consistently and well.
Anthropic Builds an Enterprise AI Services Company
This week, Anthropic announced a new enterprise AI services company built in partnership with Blackstone, Hellman & Friedman, and Goldman Sachs. The new entity will deploy and manage Claude-based AI systems for large enterprises — effectively bringing a managed-services layer to Anthropic’s technology stack.
Separately, Anthropic disclosed that its annualized revenue run rate has surpassed $30 billion, up from roughly $9 billion at the end of 2025. The company also expanded its compute partnership with Google and Broadcom to absorb the infrastructure demand that comes with that growth.
This consolidation matters for the SMB market even if you’re not buying from Anthropic directly. Managed enterprise AI services set pricing expectations, capability benchmarks, and implementation patterns that filter down to smaller-business tools within 12–18 months.
What this means for your business: Pay attention to what enterprise clients are getting from AI today — it’s typically 18 months ahead of what hits the SMB tooling market. Managed AI agents for customer service, legal review, and financial modeling are live at the enterprise level right now. If your industry includes large competitors, assume they are already deploying these capabilities and plan accordingly.
The Takeaway
This week’s arc is consistent: AI is becoming infrastructure, and governments and businesses alike are treating it that way. Pre-deployment testing, industry-wide price cuts, majority SMB adoption, and a new managed-services layer for enterprises all point to the same shift — AI is no longer a pilot program, it’s an operating assumption.
For Houston businesses trying to figure out where to start or what compliance looks like in their specific industry, BlueHill can help. We work directly with SMBs to cut through the noise and build an AI approach that fits their size, budget, and the regulations coming their way.