The bipartisan House AI Task Force is turning up the heat. Their latest target? American companies quietly plugging Chinese-built artificial intelligence into their operations.
Sources familiar with the investigation tell me the committee has sent letters to at least seven Fortune 500 firms, demanding detailed disclosures about their use of models from DeepSeek, Baidu, and other Chinese developers. The letters — sent over the past three weeks — cite concerns over data privacy, national security, and potential backdoors. The companies have until August 1 to respond.
This isn't a fishing expedition. The task force has evidence that several U.S. corporations, including a major healthcare provider and a defense contractor, have been running Chinese AI models on internal servers without fully vetting the code.
One staffer on the committee put it bluntly: 'You can't just download a black box from Beijing and plug it into your supply chain. That's how secrets leak.'
The DeepSeek Problem
At the center of the storm is DeepSeek, the Chinese startup that exploded onto the scene last year. Its models — which rival OpenAI and Google on certain benchmarks — are open-source and free. That's precisely the problem.
'American companies love free stuff,' says Mariana Torres, a cybersecurity researcher at Stanford. 'But free AI from China comes with invisible strings. The training data, the model weights, the pipeline — all controlled by a company that answers to Beijing.'
DeepSeek has repeatedly stated its models are safe and transparent. But the House panel isn't buying it. They've subpoenaed internal communications from DeepSeek's U.S. subsidiary and are pressing for a full audit of the company's data handling practices.
The timing matters. Just last month, a Department of Defense report flagged DeepSeek's models as potentially vulnerable to adversarial manipulation — a nice way of saying they could be tricked into spilling classified information.
'You can't just download a black box from Beijing and plug it into your supply chain. That's how secrets leak.'
Why Now? The AI Arms Race Heats Up
Let's be honest: Washington has been asleep at the wheel on AI regulation for years. But 2026 is different. China is no longer just copying Western tech — it's innovating. DeepSeek's latest model, released in March, beat GPT-4 on several reasoning tests. That got Congress's attention.
Meanwhile, the Commerce Department's export controls on advanced chips have failed to slow China's progress. Chinese AI developers have simply gotten more efficient with less powerful hardware. The result? Cheap, capable models that are impossible to ignore — and tempting to use.
'Export controls were supposed to strangle China's AI ambitions,' says James Harlow, a former CIA analyst who now advises tech companies. 'Instead, they accelerated their optimization. Now we have Chinese models that are good enough for most tasks, and American companies are using them to cut costs. It's a national security nightmare.'
The healthcare provider I mentioned earlier? They're using DeepSeek to analyze medical imaging. A defense contractor is using it to summarize intelligence reports. Neither practice is illegal — yet. But the committee is considering legislation that would ban the use of Chinese AI models in critical infrastructure and defense.
The Data Graveyard
Here's what really keeps lawmakers up at night: every query sent to a Chinese AI model is data that leaves U.S. soil. Even if the model is running on a local server, the underlying training data flows from China. That means every conversation, every prompt, every algorithm update is potentially visible to the Chinese government.
And it's not just government secrets. Corporate trade secrets, customer information, intellectual property — it all funnels through systems that U.S. intelligence agencies can't fully vet.
'We're building a data graveyard under the ocean between the U.S. and China,' says Torres. 'Once it's in there, you can't get it back.'
The committee is also looking at whether Chinese AI models contain hidden 'backdoors' — code that allows Beijing to access systems remotely. So far, no smoking gun. But the panel's technical advisors have found 'anomalous behavior' in some model outputs that warrant further investigation.
What Happens Next
The seven companies have until August 1 to respond. The committee has subpoena power and isn't afraid to use it. If the responses are insufficient, expect public hearings this fall. And expect the CEOs of those companies to sweat under the lights.
In the meantime, the White House is working on an executive order — expected within weeks — that would require all federal contractors to disclose any use of foreign AI models and undergo a security review. That order will likely ripple through the private sector, forcing companies to choose between Chinese AI's cost advantages and the risk of federal scrutiny.
This investigation is just the opening salvo. The real battle — over whether to ban, restrict, or embrace Chinese AI — is only beginning. And one thing is certain: the era of ignoring where your AI comes from is over.
