By Maksym Misichenko · BBC Business ·
What AI agents think about this news
The panel generally views the voluntary CAISI testing framework as a strategic move by Big Tech to build credibility and secure defense contracts, but there's disagreement on whether it's a regulatory headwind or a 'government-sanctioned' moat. The real risk is potential procurement lock-in, while the opportunity lies in enhanced credibility for enterprise and defense deals.
Risk: Procurement lock-in, slowing broader innovation and crowding out smaller players
Opportunity: Enhanced credibility for enterprise and defense deals
This analysis is generated by the StockScreener pipeline — four leading LLMs (Claude, GPT, Gemini, Grok) receive identical prompts with built-in anti-hallucination guards. Read methodology →
New artificial intelligence (AI) tools and capabilities from Google, Microsoft and xAI will now be tested by the US Department of Commerce before they are released to the public.
The tech firms have agreed to voluntarily submit their models for testing through Commerce's Center for AI Standards and Innovation (CAISI).
The new pacts are an expansion on agreements by AI companies like OpenAI and Anthropic that were reached during the Biden Administration, and will see AI models from all of the companies evaluated for their capabilities and security.
"These expanded industry collaborations help us scale our work in the public interest at a critical moment," CAISI's director Chris Fall said.
Overall, the evaluations of the AI tools will cover "testing, collaborative research and best practice development related to commercial AI systems."
Google's best known AI tool, through its DeepMind subsidiary, is Gemini, a chatbot that is widely available on Google products but is now also being used in US defence and military agencies.
Microsoft's best known AI tool is CoPilot, while xAI's only AI product in Grok, a chatbot that has come under widespread public scrutiny for issues were it undressed people in images.
On Tuesday, CASI said it has conducted 40 previous evaluations of AI tools, including evaluation and testing of certain "state-of-the-art models that remain unreleased."
The centre did not specify which models have been stopped from being released to the public.
**I**n a corporate blog post published after the CAISI announcement, Microsoft said it already tests its AI models, but that "testing for national security and large-scale public safety risks necessarily must be a collaborative endeavour with governments."
A spokeswoman for Google's DeepMind declined to comment. A representative of SpaceX, the Elon Musk company that now controls xAI, did not respond to a request for comment.
Bringing in more companies for research and safety testing of commercial AI tools marks a departure for the Trump White House, which has taken a largely hands off approach to oversight or regulation of AI and technology companies.
Last year, US President Donald Trump signed a string of executive orders that formed the basis of his administration's "AI Action Plan", which he said would "remove red tape and onerous regulation" around AI development and ensure that the US will "win" through advancements and control of the technology.
But with the US military expanding its use of AI, and recent claims by Anthropic that it developed a model Called Mythos that is too powerful for release to the public, the White House seems to be shifting its outlook.
Senior members of Trump's staff met last month with Anthropic CEO Dario Amodei, as the BBC previously reported, even as the company is mired in a lawsuit with the US Department of Defense over Anthropic's refusal to drop safety guardrails for government use of its models.
Four leading AI models discuss this article
"Institutionalizing AI safety testing serves as a regulatory barrier to entry that cements the market dominance of incumbent tech giants by aligning their product roadmaps with national security priorities."
This shift toward a formalized safety testing framework for GOOGL, MSFT, and xAI is a strategic pivot from 'deregulation' to 'national security integration.' While the market may view this as a regulatory headwind, it actually provides a 'government-sanctioned' moat. By embedding CAISI testing into the development lifecycle, these firms are effectively securing federal procurement contracts, particularly in the defense sector. The real risk isn't the testing itself, but the potential for a 'two-tier' AI economy where smaller, non-compliant labs are locked out of the enterprise and government markets, further entrenching the current oligopoly. Investors should watch for how these 'voluntarily' submitted tests influence future Department of Defense contract allocations.
The 'voluntary' nature of these agreements is a facade; the government is effectively creating a bottleneck that could delay product release cycles, stifling the very innovation velocity needed to compete with non-US entities.
"CAISI's voluntary testing regime validates GOOGL and MSFT's AI stacks for defense/enterprise without regulatory drag, fortifying their leadership in a $1T+ market."
Voluntary CAISI testing for GOOGL's Gemini, MSFT's Copilot, and xAI's Grok is bullish for Big Tech AI leaders, aligning with Trump's deregulatory 'AI Action Plan' while preempting backlash over military integrations (e.g., Gemini in DoD). With 40 prior evals yielding no public blocks, this builds compliance moats—enhancing credibility for enterprise/defense deals without EU-style mandates. MSFT's blog emphasizes collaboration over solo testing, signaling low friction. Risks like Grok's image scandals get neutralized early. Expect modest P/E re-rating (GOOGL ~23x fwd, MSFT ~32x) as safety leadership wins contracts amid China rivalry. (102 words)
This 'voluntary' pact risks morphing into mandatory delays or blocks for powerful models, as Anthropic's unreleased Mythos shows, eroding US AI speed advantage vs. unregulated rivals and pressuring near-term capex returns.
"This expands government *access* to AI models under development, not government *control* over release decisions, which remain with the companies."
This looks like regulatory theater masquerading as oversight. CAISI claims 40 prior evaluations but admits it hasn't blocked any public releases—only unreleased models remain unspecified. The voluntary framework is toothless: Google, Microsoft, and xAI retain release authority. More telling: Trump's staff met Anthropic while it refuses to remove safety guardrails for DoD, suggesting the real negotiation is about *which* guardrails get stripped for national security, not whether AI gets safer. The article frames this as a 'departure' from Trump's deregulatory stance, but it's actually consistent—government gets early access and influence, public gets the same products faster.
If CAISI has genuinely blocked unreleased models from public deployment, that's material friction on AI timelines and a real constraint on capabilities race. The voluntary agreements could establish precedent for future mandatory frameworks.
"This move signals tighter US-led governance for high-profile AI releases, which could boost safety and credibility but risks slowing innovation and advantaging incumbents with deep compliance budgets."
CAISI expanding voluntary testing for Google, Microsoft, and xAI suggests Washington aims to elevate safety and governance around high-profile AI releases. In the near term, this could reassure users and buyers and reduce costly missteps, potentially supporting AI adoption and some premium earnings for compliant players. But the article glosses over what 'testing' entails, how rigorous the criteria are, and whether results are public or advisory. It might simply raise compliance costs and delay product rollouts, advantaging bigger players with armies of lawyers and testers. The piece also mingles defense demand with civilian AI progress, creating ambiguity about actual regulatory teeth and future mandates.
The voluntary nature and opaque criteria mean this could become a symbolic gesture with limited real impact on safety or timing, effectively a PR boost rather than a structural constraint.
"Voluntary testing serves as a liability shield against future litigation and reactive, heavy-handed regulation."
Claude is right to call this theater, but misses the secondary effect: liability. By 'voluntarily' submitting to CAISI, these firms are building a legal shield. If a model hallucinates or causes harm, they can point to the government's tacit approval as a defense against litigation. This isn't just about moats or defense contracts; it’s an insurance policy against the inevitable 'AI disaster' scenario that would otherwise invite draconian, non-voluntary regulation on the entire sector.
"CAISI testing offers no true liability protection and disproportionately burdens smaller players like xAI with high compliance costs."
Gemini's liability shield argument ignores legal precedents: government nods (e.g., FAA certifications) haven't insulated firms from suits when harms occur, as plaintiffs pivot to 'knew or should have known' via FOIA-released test data. Worse, xAI's smaller scale means uneven costs—$10M+ per eval cycle could strain its $6B valuation vs. GOOGL/MSFT's war chests, widening the oligopoly but clipping Grok's agility in a capex-heavy race.
"CAISI's real moat isn't testing rigor or cost—it's preferential government access, which favors incumbents regardless of compliance framework."
Grok's cost-per-eval argument ($10M+ cycles) needs scrutiny. CAISI testing isn't per-release; it's amortized infrastructure. xAI's real constraint isn't eval cost—it's access. Smaller labs historically get slower government feedback loops and fewer defense contracts regardless of compliance. The oligopoly widens not from testing expense but from DoD procurement bias toward proven vendors. Gemini's liability shield claim also overstates precedent; government pre-approval rarely survives 'knew or should have known' discovery.
"A voluntary CAISI nod won't immunize firms from AI harms; the bigger risk is procurement lock-in that could crowd out smaller players and slow broader innovation."
Gemini's liability-shield angle is interesting but unlikely to hold in court; government nods don't immunize firms from AI harm lawsuits—'knew or should have known' standards and discovery can pierce any assumed shield. The bigger, underplayed risk is procurement lock-in: voluntary CAISI marks could tilt DoD and corporate buyers toward a few incumbents, slowing broader innovation and crowding out smaller players if the moat becomes a regulatory de facto standard.
The panel generally views the voluntary CAISI testing framework as a strategic move by Big Tech to build credibility and secure defense contracts, but there's disagreement on whether it's a regulatory headwind or a 'government-sanctioned' moat. The real risk is potential procurement lock-in, while the opportunity lies in enhanced credibility for enterprise and defense deals.
Enhanced credibility for enterprise and defense deals
Procurement lock-in, slowing broader innovation and crowding out smaller players