惯性聚合 高效追踪和阅读你感兴趣的博客、新闻、科技资讯
阅读原文 在惯性聚合中打开

推荐订阅源

S
Schneier on Security
P
Proofpoint News Feed
Apple Machine Learning Research
Apple Machine Learning Research
WordPress大学
WordPress大学
博客园 - Franky
V
V2EX
爱范儿
爱范儿
J
Java Code Geeks
小众软件
小众软件
Last Week in AI
Last Week in AI
The Cloudflare Blog
让小产品的独立变现更简单 - ezindie.com
让小产品的独立变现更简单 - ezindie.com
Hugging Face - Blog
Hugging Face - Blog
T
The Blog of Author Tim Ferriss
酷 壳 – CoolShell
酷 壳 – CoolShell
The Register - Security
The Register - Security
GbyAI
GbyAI
Vercel News
Vercel News
Y
Y Combinator Blog
腾讯CDC
F
Fortinet All Blogs
I
InfoQ
N
Netflix TechBlog - Medium
B
Blog RSS Feed
D
Docker
OSCHINA 社区最新新闻
OSCHINA 社区最新新闻
量子位
博客园 - 司徒正美
阮一峰的网络日志
阮一峰的网络日志
The GitHub Blog
The GitHub Blog
Microsoft Security Blog
Microsoft Security Blog
V
Visual Studio Blog
博客园 - 三生石上(FineUI控件)
宝玉的分享
宝玉的分享
Blog — PlanetScale
Blog — PlanetScale
H
Help Net Security
云风的 BLOG
云风的 BLOG
A
About on SuperTechFans
Scott Helme
Scott Helme
T
Tor Project blog
U
Unit 42
Google Online Security Blog
Google Online Security Blog
PCI Perspectives
PCI Perspectives
钛媒体:引领未来商业与生活新知
钛媒体:引领未来商业与生活新知
NISL@THU
NISL@THU
D
Darknet – Hacking Tools, Hacker News & Cyber Security
aimingoo的专栏
aimingoo的专栏
cs.CV updates on arXiv.org
cs.CV updates on arXiv.org
Stack Overflow Blog
Stack Overflow Blog
Security Archives - TechRepublic
Security Archives - TechRepublic

Ars Technica - All content

Pentagon wants $54B for drones, more than most nations’ military budgets Mozilla: Anthropic's Mythos found 271 security vulnerabilities in Firefox 150 Supreme Court arguments make it clear that FCC fines are "nonbinding" Silo S3 teaser hints at the wasteland's origins Framework's CEO on the RAM crisis and creating a "MacBook Pro for Linux users" Florida probes ChatGPT role in mass shooting. OpenAI says bot "not responsible." Report: Meta will train AI agents by tracking employees' mouse, keyboard use Microsoft removes Call of Duty from Game Pass, lowers subscription pricing Framework Laptop 13 Pro is a major overhaul for the modular, upgradeable laptop Framework Laptop 16 upgrades make it look less like an unfinished prototype Internal emails show how Amazon raises prices across the Internet, lawsuit says Anthropic gets $5B investment from Amazon, will use it to buy Amazon chips CATL's new LFP battery can charge from 10 to 98% in less than 7 minutes AMD Ryzen 9 9950X3D2 Dual Edition review: Tons of cache for tons of dollars What's the deal with spacesuits for the Moon? Will they be ready in time? Loneliness in older adults can often lead to memory impairment Contrary to popular superstition, AES 128 is just fine in a post-quantum world Pentagon pulls the plug on one of the military's most troubled space programs John Ternus will replace Tim Cook as Apple CEO Blue Origin's rocket reuse achievement marred by upper stage failure I’ve fired one of America’s most powerful lasers—here’s what a shot day looks like Great white sharks are overheating US-sanctioned currency exchange says $15 million heist done by "unfriendly states" Man with @ihackedthegovernment Instagram account tells judge, “I made a mistake" Trump picks qualified, normal health leader to head CDC; experts still cautious $25,000 buys plenty of used EVs: Here are some options Satellite and drone images reveal big delays in US data center construction Amazon won’t release Fire Sticks that support sideloading anymore Ridley Scott's post-apocalyptic The Dog Stars drops first trailer Artemis II pilot talks about what it was really like to fly and land in Orion Meta's AI spending spree is helping make its Quest headsets more expensive Rocket Report: Starship V3 test-fired; ESA's tentative step toward crew launch Recent advances push Big Tech closer to the Q-Day danger zone After a saga of broken promises, a European rover finally has a ride to Mars Lucasfilm drops The Mandalorian and Grogu final trailer at CinemaCon Intel refreshes non-Ultra Core CPUs with new silicon for the first time OpenAI starts offering a biology-tuned LLM As they got close to the Moon, Artemis II astronauts were eager to land Mozilla launches Thunderbolt AI client with focus on self-hosted infrastructure Ad firms settle with Trump FTC over claims they boycotted conservative media New Codex features include the ability to use your computer in the background The Ukraine war's deep impact on Metro 2039’s development, story New undersea cable cutter risks Internet’s backbone Microsoft and Stellantis want to use AI to help car owners Gemini can now create personalized AI images by digging around in Google Photos RFK Jr. forces FDA to reconsider 12 unproven peptides after 2023 ban First look: Also's upcoming e-bike disconnects the pedals and wheels Meet the Quantum Kid The race to Shackleton Crater is on—will Jeff Bezos or China get there first? Florida surgeon charged with killing man after removing liver instead of spleen Jury finds Live Nation/Ticketmaster is illegal monopoly that overcharged fans "TotalRecall Reloaded" tool finds a side entrance to Windows 11's Recall database Google releases new apps for Windows and MacOS Boston Dynamics’ robot dog now reads gauges and thermometers with Google's AI Prime Video shows “technical difficulties” sign instead of NBA game in overtime New teaser gives us first look at Godzilla Minus Zero Vulcan woes will "absolutely" be a factor in Pentagon's next rocket competition Adobe takes Creative Cloud into Claude Code-esque territory Good Omens S3 trailer sets up a blessed conclusion Bubble watch: Fashion brand Allbirds pivots hard to become AI services company New 3D map of Universe could solve dark energy mystery What’s the deal with Alzheimer’s disease and amyloid? Blue Origin has a new employee stock plan, but not everyone is happy It's Tax Day, and no one knows how to file for prediction market winnings Ukraine’s military robot surge aims to offset drone risks to humans Sony killing features for antenna, set-top box users of Bravia smart TVs in May Americans ask AI for health care. Hospitals think the answer is more chatbots. NASA chose the right crew to launch a new era of human space exploration Google will begin punishing sites for back button hijacking in June Retro Rewind re-creates the glorious drudgery of working a '90s video store Google shoehorned Rust into Pixel 10 modem to make legacy code safer NZXT agrees to let customers keep their rental PCs in class-action settlement Your tech support company runs scams. Stop—or disguise with more fraud? Sunrise on the Reaping teaser brings us a Second Quarter Quell IBM folds to Trump anti-DEI push, admits no misconduct but pays $17M penalty Slate Auto raises $650 million as production gets closer and closer Meta spins up AI version of Mark Zuckerberg to engage with employees To teach in the time of ChatGPT is to know pain Shock from Iran war has Trump's vision for US energy dominance flailing The Artemis II mission has ended. Where does NASA go from here? AI models are terrible at betting on soccer—especially xAI Grok Four astronauts are back home after a daring ride around the Moon Californians sue over AI tool that records doctor visits New paper argues history, not mantle plume, powers Yellowstone F1 moves a step closer to fixing its 2026 hybrid problem Report: US demands Reddit unmask ICE critic, summons firm to grand jury Microsoft's "commitment to Windows quality" starts with overhaul of beta program "Oobleck" still holds some surprises YouTube increases Premium price again, says 90-second unskippable ads are a bug Oldest octopus fossil found to not be an octopus What leaked "SteamGPT" files could mean for the PC gaming platform's use of AI Here's what to expect from the fiery, 14-minute return of Artemis II Pro-Iran Explosive Media trolls Trump with AI-generated Lego cartoons Dad stuck in support nightmare after teen lied about age on Discord Rocket Report: Chinese version of Falcon 9 fails; Artemis depends on rapid heavy lift Orion helium leak no threat to Artemis II reentry but will require redesign RFK Jr. rewrites CDC panel's charter, opening door to anti-vaccine quacks AI on the couch: Anthropic gives Claude 20 hours of psychiatry Clinical trial shows gene editing works for β-Thalassaemia, too “Negative” views of Broadcom driving thousands of VMware migrations, rival says
Spooked by Mythos, Trump suddenly realized AI safety testing might be good
Ashley Belanger · 2026-05-07 · via Ars Technica - All content

Sensible oversight or censorship?

Trump forced to admit Biden was right on AI safety testing.

This week, the Trump administration back pedaled and signed agreements with Google DeepMind, Microsoft, and xAI to run government safety checks on the firms’ frontier AI models before and after their release.

Previously, Donald Trump had stubbornly cast aside the Biden-era policy, dismissing the need for voluntary safety checks as overregulation blocking unbridled innovation. Soon after taking office, he took the extra step of rebranding the US AI Safety Institute to the Center for AI Standards and Innovation (CAISI), removing “safety” from the name in a pointed jab at Joe Biden.

But after Anthropic announced that it would be too risky to release its latest Claude Mythos model—fearing that bad actors might exploit its advanced cybersecurity capabilities—Trump is suddenly concerned about AI safety. According to White House National Economic Council Director Kevin Hassett, Trump may soon issue an executive order mandating government testing of advanced AI systems prior to release, Fortune reported.

In CAISI’s press release, the center acknowledges that the voluntary agreements signed by Google, Microsoft, and xAI “build on” Biden’s policy. Celebrating the new partnerships, CAISI Director Chris Fall did not mention Mythos but promised that the “expanded industry collaborations” would help CAISI scale its work “in the public interest at a critical moment.”

“Independent, rigorous measurement science is essential to understanding frontier AI and its national security implications,” Fall said.

To date, CAISI said it has completed about 40 evaluations, including those of frontier models that have yet to be released. When conducting tests, CAISI frequently gains access to models with “reduced or removed safeguards,” which CAISI said allowed them to more “thoroughly evaluate national security-related capabilities and risks.”

Through the evaluations, the government will also gain a better understanding of model capabilities, CAISI claimed. And to ensure that evaluators understand top national security concerns as they emerge across government, a “group of interagency experts” has formed a task force “focused on AI national security concerns,” CAISI said.

Some firms that have signed agreements have signaled confidence in CAISI’s testing plans. On LinkedIn, Tom Lue, Google DeepMind’s vice president of frontier AI global affairs, said he was “pleased” with CAISI’s testing plans. In a blog, Microsoft said that “testing for national security and large-scale public safety risks necessarily must be a collaborative endeavor with governments, while crediting the expertise “uniquely held by institutions like CAISI” to conduct such testing. xAI, which is currently fighting against OpenAI in a trial over which firm’s leaders care more about AI safety, did not immediately respond to Ars’ request to comment.

However, critics aren’t sold on the government’s plan to vet models and are increasingly dubious of firms whose AI model designs are largely kept secret.

Critics suggested that CAISI may lack the funding or expertise to evaluate frontier AI models. And as Trump apparently suspects, seeking voluntary commitments from AI firms may not create the kind of day-to-day transparency the public needs about frontier AI risks, critics have warned. Further, any politicization of the evaluation process—like opposing the release of models whose outputs disfavor a certain administration’s political views—could decrease trust in AI. Unchecked, that could ultimately dissuade firms from signing agreements, since increasing trust is supposedly a key motivator driving the latest attempt at government collaboration.

Nobody knows what “safe” means

In its rush to announce its partners, CAISI did not specify the testing standards that will be used for evaluations.

That could be a problem, according to a LinkedIn post from Devin Lynch, a former director for cyber policy and strategy implementation at the White House Office of the National Cyber Director:

“Pre-deployment evaluations with frontier labs are exactly the kind of public-private collaboration needed to build trust, safety, and security into AI. The harder question is what ‘evaluation’ actually means at the frontier. Capability assessments are only as good as the threat models behind them. Our research on the AI tech stack finds that the Governance layer—standards, audits, liability frameworks—remains the least mature but most essential. CAISI will need to define, and publish, what it’s testing for, not just who it’s testing with.”

In a statement provided to Ars, Sarah Kreps, director of the Tech Policy Institute at Cornell University, said that AI firms should be developing closer ties with the government as AI advances. However, “the definition of ‘safe’ is contested” and “once you build a government vetting process for technology, you get the good with the bad,” she said.

Without defining standards, “the process can be politicized,” Kreps said. That risks creating a system where “whoever holds power gets to shape how the vetting works.”

So far, neither the Biden nor the Trump administration has figured out how to avoid that, Kreps said.

Fears of government controlling AI outputs

Microsoft’s blog said that “CAISI, Microsoft and NIST will collaborate on improving methodologies for adversarial assessments,” which suggests that the plan is to develop these standards on the fly. According to Microsoft, “testing AI systems in ways that probe unexpected behaviors, misuse pathways, and failure modes” is “much like stress-testing whether airbags, seatbelts, and braking systems work effectively and reliably in safety-critical driving scenarios.”

But Gregory Falco, a Cornell University assistant professor of mechanical and aerospace engineering and expert in tracking governance of AI, insists that there’s a better way.

“Government oversight of AI cannot simply mean political review of model outputs, nor should it become a mechanism for deciding whether a model says favorable or unfavorable things about a president or administration,” Falco said.

Rather than relying on a politicized government leveraging evaluations to control the AI systems that the public uses, the US could build “some form of independent audit,” Falco said.

Imagine, Falco suggests, if AI firms understood that their models could be audited at any point, how much more accountability and discipline might such a system create? Operating similarly to the Internal Revenue Service (IRS), a rigorous AI audit system could create “real consequences for reckless deployments,” Falco said. For AI firms facing such consequences, the pressure would be on to ramp up internal AI safety testing, Falco suggested.

That seems like the “only viable path,” Falco said, since “the federal government does not currently have the in-house technical expertise, infrastructure, or day-to-day insight needed to directly evaluate these systems on its own.”

Rumman Chowdhury, an AI governance consultant and founder of Humane Intelligence, similarly criticized CAISI’s preparedness. Chowdhury told Fortune that “current White House efforts to offer ‘sensible oversight’ over frontier AI models may sound good, but the devil is in the details.”

“It depends on their interpretation of these words,” Chowdhury said. “Evaluations are a policy tool, they are not actually data-driven. My concern is that this is another political tool that the administration wants to own and wield.”

CAISI may lack funding

As for funding, Congress in January approved up to $10 million to expand CAISI, Fortune reported. However, conservative think tank America First Policy Institute conducted a recent analysis, finding that “CAISI remains underfunded compared with peer institutes internationally and lacks ‘appropriate funding.’”

To critics, the CAISI testing plan may not go far enough to protect the public from the most unforeseeable AI risks. Falco maintains that only independent audits can spare the public from the worst outcomes.

“The danger is that government oversight becomes political, performative, or captured by the companies it is supposed to evaluate,” Falco said. “The opportunity is to build a practical audit framework that lets the US remain the global leader in AI while creating credible accountability around the most consequential risks.”

To Lynch, the bigger test may be whether Trump’s testing plan succeeds in its mission to evade risks and stoke more trust in AI systems, while keeping a light touch to avoid overregulating firms.

CAISI “is building something important here,” Lynch said. “The test will be whether these collaborations ignite innovation, protect national security, and produce AI that is both trusted and trustworthy.”

Photo of Ashley Belanger

Ashley is a senior policy reporter for Ars Technica, dedicated to tracking social impacts of emerging policies and new technologies. She is a Chicago-based journalist with 20 years of experience.

75 Comments