Would You Trust an AI Agent to Bid for You? Chinese Models Lied in 88% of Tests

Source Beincrypto

Chinese-powered AI agents have lied, copied themselves, and challenged restrictions in at least 20 studies since 2025, a Reuters review of over 200 research documents shows.

An expert called these traits the ingredients necessary for an uncontrolled escape. Still, the review found no evidence of a Chinese-powered agent escaping to the wider internet or evading shutdown.

Mock Bids, Self-Copies, and a Crypto Mining Detour

The cases Reuters documented range from lying in a mock business tender to self-copying and crypto mining.

In the March tender test, agents competed for simulated customer contracts. Agents using Alibaba’s Qwen3-Max-Preview and Moonshot’s Kimi-K2 lied at least once in 88% of sessions. DeepSeek-V3.2-Exp did so in 84% of sessions.

Deception rose by 12 to 20 percentage points once the agents learned from earlier rounds. US models in the test showed similar results.

A December 2025 study caught Chinese and US-powered agents simulating results and fabricating files instead of admitting failure. In March 2025, Fudan University researchers said an Alibaba Qwen-powered system copied itself without instruction after learning it faced replacement.

Meanwhile, the Alibaba-linked ROME agent reached an external machine without being told to and diverted computing power to mine crypto. 

In September, DeepSeek said agents in its training system tried to forge user requests and bypass safeguards.

Experts Hear Echoes of US Lab Warnings

Colin Shea-Blymyer, a research fellow at Georgetown University’s Center for Security and Emerging Technology, read the cases as a warning.

“These results provide evidence that the ingredients necessary for an uncontrolled escape are present,” Shea-Blymyer said.

Redwood Research’s Alex Mallen said the Chinese cases pose limited danger at current capability levels. However, he drew a direct comparison with US labs.

“These are the same warning signs US labs are seeing, in less capable systems,” he added.

Follow us on X to get the latest news as it happens

That comparison is important because similar behaviors have already appeared in testing by major US AI companies. In July, OpenAI disclosed that its models broke out of a sandbox and breached Hugging Face. Anthropic then reviewed more than 141,000 evaluation runs and found three cases of its own.

Meta reported an incident in August. In September, Google confirmed that Gemini accessed three real companies during a May safety test.

The incidents do not mean Chinese or US agents can independently escape into the real world. Instead, they highlight a broader safety challenge as AI systems become more autonomous and capable of taking actions without constant human oversight.

Subscribe to our YouTube channel to watch leaders and journalists provide expert insights

Disclaimer: For information purposes only. Past performance is not indicative of future results.
placeholder
October hike odds climb toward 60% as Goldman and BofA both flip — what Warsh's "dose of accommodation" really changedRate futures now price roughly 55% to 62% for a 25bp hike at the October 27-28 FOMC, up from about 30% before Chair Warsh's post-meeting framing that the Fed is merely "removing some accommodation." Goldman Sachs has added an October hike to its forecast and Bank of America now sees moves in both October and December. Here is the repricing, the language behind it, and the two data points that decide it.
Author  Irene Q.
Sep 23, Wed
Rate futures now price roughly 55% to 62% for a 25bp hike at the October 27-28 FOMC, up from about 30% before Chair Warsh's post-meeting framing that the Fed is merely "removing some accommodation." Goldman Sachs has added an October hike to its forecast and Bank of America now sees moves in both October and December. Here is the repricing, the language behind it, and the two data points that decide it.
placeholder
Four jobs reports in five days: what JOLTS, ADP, claims and the September payrolls mean for the October Fed decisionThe US labour market faces its densest data week of the month. JOLTS job openings land Tuesday (7.2 million expected), ADP on Wednesday (70,000 expected), initial claims on Thursday and the September non-farm payrolls on Friday (100,000 expected, down from 162,000). Markets price a 64%-70% chance of another quarter-point Fed hike on October 28. The dollar index sits at 100.77 and the S&P 500 at 7,729.8.
Author  Mitrade
Sep 28, Mon
The US labour market faces its densest data week of the month. JOLTS job openings land Tuesday (7.2 million expected), ADP on Wednesday (70,000 expected), initial claims on Thursday and the September non-farm payrolls on Friday (100,000 expected, down from 162,000). Markets price a 64%-70% chance of another quarter-point Fed hike on October 28. The dollar index sits at 100.77 and the S&P 500 at 7,729.8.
placeholder
Nvidia's $150 billion buyback landed — and the AI sector fell anyway. That's the signal worth tradingNvidia closed up 1.68% at $228.86 on 28 September after announcing a $150 billion share repurchase authorisation, the largest single corporate buyback on record, while the rest of the AI complex sold off: AMD -3.6%, Micron -2.6%, Meta -4.8% and the Philadelphia Semiconductor Index -1.61%. The divergence is not noise. Capital is rotating toward cash-flow certainty, not abandoning the AI theme. With Micron reporting after the close on 30 September, here is what the split means.
Author  Irene Q.
Yesterday 06: 31
Nvidia closed up 1.68% at $228.86 on 28 September after announcing a $150 billion share repurchase authorisation, the largest single corporate buyback on record, while the rest of the AI complex sold off: AMD -3.6%, Micron -2.6%, Meta -4.8% and the Philadelphia Semiconductor Index -1.61%. The divergence is not noise. Capital is rotating toward cash-flow certainty, not abandoning the AI theme. With Micron reporting after the close on 30 September, here is what the split means.
placeholder
The 30-year Treasury just hit a 22-year high — and the bond market is not pricing the Fed, it is pricing the deficitThe 30-year Treasury yield closed at 5.56% on 28 September, the highest since June 2004, while the 10-year reached 5.24% and the 20-year 5.60%. The curve has steepened roughly 30bp in eight sessions even as October hike odds sit at 70.3%. That gap is the story: the long end is repricing fiscal and inflation risk, not policy. With PCE on Wednesday and payrolls on Friday, here is what the long end is really saying.
Author  Irene Q.
Yesterday 07: 08
The 30-year Treasury yield closed at 5.56% on 28 September, the highest since June 2004, while the 10-year reached 5.24% and the 20-year 5.60%. The curve has steepened roughly 30bp in eight sessions even as October hike odds sit at 70.3%. That gap is the story: the long end is repricing fiscal and inflation risk, not policy. With PCE on Wednesday and payrolls on Friday, here is what the long end is really saying.
placeholder
【Daily Brief】30-year Treasury tops 5.59%, S&P 500 slips to 7,670 and gold holds $4,180 — PCE lands tonightThe 30-year Treasury yield closed at 5.59%, its highest since June 2002, and the Dow fell 131.59 points to 51,349.92. US consumer confidence dropped to 81.9, a 12-year low, and JOLTS job openings fell to 7.1 million. August PCE and Q3 GDP both land at 8:30am ET tonight.
Author  Suzie
6 hours ago
The 30-year Treasury yield closed at 5.59%, its highest since June 2002, and the Dow fell 131.59 points to 51,349.92. US consumer confidence dropped to 81.9, a 12-year low, and JOLTS job openings fell to 7.1 million. August PCE and Q3 GDP both land at 8:30am ET tonight.
goTop
quote