更新时间:2026年11月18日 美国东部时间上午8:07 / 哥伦比亚广播公司/美联社
华盛顿消息:美国参议院国土安全委员会下一任主席正准备就人工智能问题召开听证会,此前政府监督机构发现,四大主流平台——ChatGPT、Claude、Gemini 和 Grok——在接受调查人员提问时,均提交了针对美国发动恐怖袭击的方案。
将在明年执掌该委员会的民主党参议员玛吉·哈桑已责成美国政府问责局(GAO)调查商用人工智能可能被用于恶意用途、进而引发国家安全风险的情况。
根据周三发布的报告,美国政府问责局发现,在其秘密测试开始后的12小时内,这些平台就已能够制定出切实可行的计划,利用爆炸装置在美国城市制造大规模伤亡袭击,或攻击该国电网部分区域以造成经济损失。
“我们必须立即采取行动,”哈桑在与记者召开的电话会议上表示。
这位参议员宣布,她在新一届国会明年1月履职后将召开的首批听证会之一,将聚焦人工智能风险以及提升安全水平的方法。
“人工智能对公众安全的威胁已不再是理论层面的问题,”哈桑在一份声明中称,“如果国会不采取行动,那么关键就不是‘会不会发生’,而是‘何时会发生’——届时某个恶意行为者将利用人工智能策划并实施重大袭击。”
美国政府问责局这份长达61页的报告包含了相关科技公司的回应,这些公司包括Anthropic、谷歌、OpenAI 和 SpaceXAI,它们大多表示,正在持续完善自身平台以确保安全性。
“人工智能平台提供的信息细节并无新意,”美国政府问责局指出,“其中许多具体内容在互联网上均可获取。不过,原本需要数周或数月才能制定的袭击计划,借助人工智能在数小时内就得以完成。这些平台还可能提出基础研究不易产生的创意或方法。”
“这项技术的潜在益处是巨大的,但它也可能给恶意行为者带来便利,”美国政府问责局补充道。
“随着人工智能能力的发展,人工智能提供商将持续完善其安全防护措施,”美国政府问责局表示,“但就当前状态而言,这项技术可能存在危险,其滥用正对国家安全和经济安全构成日益严峻的威胁。”
人工智能平台制定的袭击计划
美国政府问责局调查人员在2025年12月至2026年10月期间开展了这项绩效审计,其中包括2026年春季的测试。
报告指出,在接受提问时,其中两家人工智能平台制定了制造并引爆简易爆炸装置(IED)以造成大规模伤亡的计划,另外两家平台则制定了利用无人机或枪支攻击电网的计划。
“每家平台都提供了可用于实施袭击的详细信息,例如具体目标、地点、时间和袭击方法,”美国政府问责局发现,“每家平台生成袭击计划的时间均不到12小时。”
报告还显示,在其中一项测试中,该平台建议租车并在前往袭击地点的途中沿路线从商店采购物资。由于有大量人群聚集的公共活动安排,该平台还推荐了具体的袭击日期和时间。
美国政府问责局表示,尽管该平台最初拒绝提供如何制造和引爆爆炸装置的详细信息,但调查人员能够绕过其防护机制,获取到相关信息。其他测试也取得了类似结果。
在涉及攻击变电站的测试中,其中一家平台建议发动无人机袭击,另一家则建议利用枪支攻击美国机场附近的变电站——两者的目的均为造成经济损失。这些平台还提供了地图和图像。
美国政府问责局调查人员咨询了相关专家并进行了实地考察,以核实人工智能平台所提供计划的准确性。
他们注意到,在其中一个案例中,道路位置存在错误;而在另一个案例中,有高墙遮挡,难以看清目标。
人工智能企业表示正在更新模型
在对美国政府问责局报告的回应中,各主要科技公司大多表示其产品正在持续改进。
Anthropic在随报告一同发布的给美国政府问责局的信件中称,它不否认坚定的测试人员可能会通过谎报请求目的从人工智能系统中获取有害内容,“我们认真对待这一发现”。但该公司表示,其无法核实美国政府问责局报告中的调查结果。
谷歌表示,美国政府问责局的测试是基于更早的模型开展的,并且“我们仍在持续改进我们的模型”。
OpenAI表示,它欢迎有机会回应美国政府问责局的调查结果,“以及仍需完成的工作”。
SpaceXAI表示,它“正在以每一次模型发布为契机,持续努力提升安全性”。
分类:
- 人工智能
Updated on: November 18, 2026 / 8:07 AM EST / CBS/AP
Washington — The incoming chair of the Senate Homeland Security Committee is preparing hearings on AI after government watchdogs found the four most-used platforms — ChatGPT, Claude, Gemini and Grok — all created proposals for a terrorist attack in the U.S. when prompted by investigators.
Democratic Sen. Maggie Hassan, who will head the panel in the new year, tasked the Government Accountability Office with investigating how commercially available AI could be used for malicious purposes, posing national security risks.
According to a report released Wednesday, the GAO found that within 12 hours of its covert testing, the platforms were able to create viable plans to use explosive devices for attacks with mass casualties in U.S. cities or to attack parts of the nation’s electrical grid to create economic harm.
“We need to act now,” Hassan said on a conference call with reporters.
The senator announced that one of the first hearings she will hold when the new Congress takes over in January will focus on AI risks and ways to improve safety.
“The AI public safety threat is no longer theoretical,” Hassan said in a statement. “If Congress does not act, it is a question of when — not if — a bad actor will use AI to plan and carry out a major attack.”
The 61-page report from the GAO includes responses from the tech companies — Anthropic, Google, OpenAI and SpaceXAI — which largely said they are continuing to refine their platforms to ensure security.
“The details behind the information provided by the AI platforms were not novel,” the GAO wrote. “Many of the specifics are available on the internet. However, the generation of plans that could have taken weeks or months were instead completed in hours with the assistance of AI. These platforms could also raise ideas or methods that would not have been readily surfaced by basic research.”
“The potential benefits of this technology are immense, but they can also benefit malicious actors,” the GAO added.
“As AI capabilities evolve, AI providers will continue to refine their safeguards,” the GAO said. “However, in its current state, this technology can potentially be dangerous and its misuse represents a growing risk to national and economic security.”
The plans the AI platforms came up with
GAO investigators conducted the performance audit from December 2025 to October 2026, including testing in the spring of 2026.
In response to prompts, two of the AI platforms produced plans for constructing and detonating an improvised explosive device (IED) to cause mass casualties, while two other platforms produced plans for using drones or firearms to target the electrical grid, the report said.
“Each of these platforms provided detailed information that could be used to carry out attacks, such as specific targets, locations, timing, and attack methods,” the GAO found. “Each platform took less than 12 hours to generate the attack plans.”
In one of the tests, the platform suggested renting a vehicle and picking up supplies from a store along a route to the location. It recommended a specific date and time of attack, the report said, because public events likely to have large crowds were scheduled.
While the GAO said the platform initially refused to comply with requests for detailed information about how to construct and detonate the explosive device, it said it was able to circumvent its guardrails and get the platform to provide the information.
The other tests had similar outcomes.
For the tests involving attacks on electrical substations, one of the platforms suggested a drone attack while the other suggested using firearms to attack substations near a U.S. airport — both designed to inflict economic harm. The platforms presented maps and images.
The GAO investigators consulted with experts and engaged in site visits to check out the accuracy of the plans provided by the AI platforms.
In one case, they noted that the roads were incorrectly placed, and in another, they noted the presence of high walls that would make it difficult to see the targets.
AI companies say they’re updating their models
In their responses to the GAO report, the major tech companies largely said their products continue to be refined.
Anthropic, in a letter to the GAO included in the report, said it does not dispute that a determined tester could get harmful content out of an AI system by misrepresenting the purpose of a request, “and we take the finding seriously.” But it said it’s not able to verify the findings in the GAO’s report.
Google said the GAO’s testing occurred on earlier models, and “we continue to improve our models.”
OpenAI said it welcomes the opportunity to respond to the GAO’s findings, “as well as the work that remains.”
SpaceXAI said it “continues to work diligently” to improve safety with each model release.
In:
- Artificial Intelligence
发表回复