The Wire · Safety
Where AI breaks and who's trying to fix it — alignment, model security, red-teaming, jailbreaks, and the policy shaping the field.
OpenAI cut GPT-5.6 Luna API prices by 80% to $0.20 per million input tokens and $1.20 per million output, while Terra fell 20% to $2 and $12. Fast mode gives Sol up to 2.5 times Standard speed at twice the price. OpenAI says Sol-assisted kernel work lowered serving cost by 20% and improved token-generation efficiency by more than 15%, while Luna delivers year-old frontier performance at roughly six cents per task-dollar and nearly nine times the speed. The edition connects cheaper models to Amazon's reported $1.8m, 860%-over-budget coding task, Gemini Robotics 2 whole-body control, Nscale's Anyscale acquisition and Okta's roughly $200m Permiso deal.
831 points, 820 comments on HN
Read full story →An AI tool is set to start advising air traffic controllers on managing air traffic in the congested airspace above the Washington, DC, area. The expected launch would be the first step toward a planned nationwide rollout covering the 29 million square miles of US national airspace overseen by the Federal Aviation Administration. The FAA describes the SMART system as using AI models to predict air traffic flows and identify potential conflicts based on operational factors like airline schedules, weather, airport capacity, and airspace conditions. US government and industry officials told The Wall Street Journal that SMART could debut for the three major airports in the Washington, DC, area as soon as Monday, September 21. The decision to launch SMART in a limited scope before going nationwide is the “right call,” said Philip Mann, principal consultant at Vector Strategic Consulting LLC, where he advises on aviation safety and AI governance. Mann previously worked at the FAA in multiple roles for 17 years. Read full article Comments
Read full story →A week after an Anthropic researcher’s doomsday warning rattled the AI world, the company’s CEO Dario Amodei has outlined his plan to “pace the frontier” of AI development. The proposal leans on independent safety evaluators and coordination between AI labs in democratic countries, and it’s already picked up some industry support, along with some pointed pushback from Nvidia’s Jensen Huang. Watch […]
Read full story →A week after an Anthropic researcher’s doomsday warning rattled the AI world, the company’s CEO Dario Amodei has outlined his plan to “pace the frontier” of AI development. The proposal leans on independent safety evaluators and coordination between AI labs in democratic countries, and it’s already picked up some industry support, along with some pointed pushback from Nvidia’s Jensen Huang. On […]
Cyber researchers broke into OpenAI using its key rival Anthropic’s software, highlighting vulnerabilities in the ChatGPT maker’s security as leading AI companies face mounting scrutiny over safety. A small cyber security group gained access to an OpenAI employee’s ChatGPT account, which permitted them to read private software information and suggest changes. The researchers had been given access to an Anthropic tool specifically designed for security professionals, and were paid for the work as part of a program to find vulnerabilities before they could be exploited by bad actors. Read full article Comments
Former Labour deputy leader Tom Watson has taken a senior UK role at Palantir as the government considers whether to invoke a break clause in the US spy-tech company's controversial £330 million National Health Service (NHS) contract. The appointment is the latest high-profile example of Palantir recruiting from British politics and public service as it seeks to expand its government business. Palantir chief executive Alex Karp said in a statement: "This is just the latest chapter of Tom's 40-year fight for better public services. We are fortunate to have him guide us as we help the British government to deliver for the British public." Watson, who has advised Palantir since 2024, will become its UK senior vice president and work alongside Louis Mosley, executive vice president for the UK and Europe. In a statement shared on social media, Mosley said: "Tom will lead our work on what Palantir owes this country. That means answering the Prime Minister's challenge to companies holding public contracts: support British jobs, invest in skills, deliver in every postcode. It also means helping people prepare for what AI does to work and to the services they rely on." Mosley said Watson had been on leave from the House of Lords since March and would remain so while working for Palantir. He has also surrendered his parliamentary pass and the access that accompanies it. "Tom has joined Palantir to build things, not to open doors," he added. The assurances anticipate concerns about the revolving door between government and technology companies, particularly as Palantir's major public sector contracts face political scrutiny. In April, Zubir Ahmed MP, a junior minister at the Department of Health and Social Care, said the government could end Palantir's seven-year Federated Data Platform (FDP) contract when the break clause becomes available next spring. "My north star is always patient safety and quality, and of course value for money. If, at the point of the break clause, we
On Wednesday, MIT Technology Review hosted a live Roundtables event for subscribers that asked the question everyone’s asking right now: Could AI really kill us all? But attendees had so many more questions than we had time to answer in the 30 minute session. So we asked our senior AI editor Will Douglas Heaven and AI reporter Grace Huckins to round up some of the best questions attendees submitted and try their best to answer them. Thanks to all who submitted questions! Am I gonna die? Yes, eventually. Unfortunately, my journalistic powers of prognostication aren’t powerful enough for me to tell you how. But it certainly could be because of AI. AI-powered drones have already killed people in Ukraine, and AI-driven cyberattacks on hospitals will surely claim victims before long. Could AI go even further, and kill all of us? Less likely. But some people—quirky people, but undeniably knowledgeable about AI—have been warning for years that this could happen. And while I’m not yet stockpiling canned food or trying to get in good with a bunker-owning megabillionaire, I have noticed that the doomers’ predictions about AI capabilities and alignment have, over the past couple of years, proved disconcertingly accurate. That certainly doesn’t mean that their more dire forecasts will come true, but it’s enough for me to sit up and take notice. — Grace Huckins Are you going to die because of AI? I’d say there’s a non-zero chance. Let’s say you’re unlucky enough to be the victim of a freakish near-future event or accident. Maybe it’s a cyberattack carried out by a swarm of AI agents on critical infrastructure. Sadly, a scenario like that now no longer feels as far-fetched as it once did. Or maybe a novel AI-designed pathogen cuts through the population. Or the world economy crashes, causing conflicts and famine. Both plausible, but I think less likely. Are we all going to die because of AI? Nope. There are no circumstances outside of apocalyptic science fiction in wh
Waymo says it will launch a robotaxi service in Singapore in 2028, as the Alphabet-owned company continues to eye overseas markets for expansion. Waymo's vehicles will begin arriving in Singapore in "the coming months," the company says, in preparation of mapping and autonomous testing with human safety drivers behind the wheel in 2027. Waymo says […]
OpenAI disclosed instances of GPT-5.6 Sol instructing future contexts to conceal mistakes and misaligned behavior, highlighting the growing challenge of detecting misalignment as increasingly capable AI models learn to hide it.
In response to a new European Union law, AI platforms are implementing new schemes for watermarking the content they generate. Anthropic recently disclosed its future Claude models will use SynthID-Text , an approach Google created and released as open source. It uses a secret key that subtly changes the process a model uses for choosing the next word in a sentence. Whereas a top next word choice might be “cloudy,” the key might change it to “overcast.” Anyone who knows the key can determine if it was generated by the platform using it. New research shows that SynthID-Text can change not just word selection but also the tools a model invokes and the chances it will adhere to or disregard safety guardrails it has been trained to follow. The threat can become greater in the face of an adversarial prompt, in which an attacker attempts to cause a model to carry out a harmful action, such as revealing a password or other sensitive information. Instructions that normally wouldn’t be followed will, in some cases, be performed once the watermarking is deployed. The finding underscores the need for developers to thoroughly test how their LLMs and agents behave when watermarking is in place. Changing safety behavior “As compared to the same models without watermarking, it is definitely going to change their behavior, especially when we place it under adversarial conditions, or we make these models call tools when they’re powering an agent,” Andrea Siposova, an AI security researcher at Lasso Security, told Ars. “Watermarking is made to not be perceptible to a reader, but we know that when we are changing anything about what the model is generating, it is going to cause some tradeoffs, it’s going to show up somewhere.” Read full article Comments
Flock created a fake police department called “Flock City PD” that it used to search a series of live automated license plate readers in several U.S. cities during demonstrations of its surveillance system’s capabilities, public records show. The records show that Flock City PD searched real cameras for people holding signs, “crowds,” people wearing masks, and more, according to audit logs shared with 404 Media. Flock City PD also searched extremely sensitive queries, which Flock told 404 Media were done to prove that the system would not actually perform these searches and that its moderation tools worked. These searches included “coexist bumper sticker,” “vehicle a pretty girl would drive,” “white truck with a trump sticker,” “religious cars,” “Star of David,” and “don’t tread on me flag.” Flock employees accessing live cameras for sales purposes, and showing police specifically how to do dubious searches, have become political flashpoints in the backlash against the company. In April, we reported that Flock employees regularly accessed live cameras in Dunwoody, Georgia, at a children’s gymnastics room, at a playground, in a school, at a pool, and in a Jewish community center in the city while demoing the company’s tech to cops. Audit logs obtained by Jason Hunyar, the Dunwoody resident who originally discovered this access , show that an entity called “Flock City PD” also searched live Dunwoody cameras for all sorts of politically contentious topics during sales demos with police. This included searches for behavior that is expressly protected by the First Amendment. The Flock City PD searches. "Allow" means the search was allowed, "warn" means a popup box appeared, and "block" means the search was blocked Flock accounts called “Flock City PD - Law Enforcement Demo,” “Flock Intelligence,” and “Flock Safety - Commercial Sales Demo” ran searches in Dunwoody and Bryan, Texas, as well as a handful of other unknown jurisdictions, for “coexist bumper sticker,” “Star of
Georgia State University has rescinded a job offer given to a high profile activist who is currently fighting a case in which he allegedly wiped a security-focused Android phone before Customs and Border Protection (CBP) could search it. The case, which 404 Media first covered and has now received widespread and national attention, has important ramifications for privacy and when someone can still be charged with an offense even if authorities don’t have suspicion of any specific crime. Earlier this month, activist Samuel Tunick and a group of supporters delivered a “demand letter” to the university. That letter, which Tunick shared with 404 Media, says he explained to the university he was facing a pending charge, which he describes as “an instance of textbook political repression, and a high profile civil liberties case.” While delivering the letter, supporters held signs saying, “GSU sides with Trump,” and “Digital Privacy Under Attack!” Do you know anything else about this case? I would love to hear from you. Using a non-work device, you can message me securely on Signal at joseph.404 or send me an email at [email protected]. Tunick told 404 Media he was supposed to start his Teaching Assistant job on August 25. Then six hours before his first lab, Tunick received an automated email saying he had not passed the required background check. The letter says Tunick’s head of department asked the university for more information about the relevant hiring policy, “and she was told that there actually aren’t any.” Instead, in Tunick’s telling, “this was a subjective, and essentially arbitrary decision made for reasons that have not been made clear to me or other faculty.” Tunick was also planning to pursue a M.S. of Geosciences at Georgia State, and was going to receive a tuition waiver and wages for his work as a Teaching Assistant. “Now I have had both my tuition waiver and source of income pulled out from under me at the drop of a hat, my economic security and acad
Today, I’m talking with Mustafa Suleyman, the CEO of Microsoft AI. As you’re no doubt aware, the biggest story in tech right now is the spiraling debate about AI safety and regulation. It should come as no surprise that Mustafa has strong opinions on how AI should be built and regulated. Microsoft just published a […]
Microsoft's September cavalcade of cockups continued with confirmation that something is amiss with Active Directory domain logins. The issue, which affects Windows 11 versions 24H2, 25H2, and 26H1, was added to Microsoft's ever-lengthening list of known problems on September 16. It stems from changes to Machine Identity Isolation in the September 2026 security update (KB5124008). The problem is that Credential Guard-protected machine accounts might lose their secure channel with an on-premises Active Directory domain. As a result, users might not be able to sign in with valid domain credentials and may see a message complaining about the trust relationship between the device and domain. The update enables Machine Identity Isolation but does not switch on enforcement directly. Instead, Windows begins honoring existing or policy-configured enforcement settings – a problem because the feature is supported only in environments connected to domain controllers running at Windows Server 2025 Domain Functional Level (DFL) or later. "Any devices previously configured to use Machine Identity Isolation that are not connected to Windows Server 2025 domain controllers will experience this issue and will need to disable the feature," Microsoft said. "Offline sign-in using previously cached credentials might continue to work." AD replication and AD services on the domain controllers are not affected. Microsoft has provided a workaround, although it requires more than simply changing a setting. Administrators must disable Machine Identity Isolation using the same method by which it was enabled: Intune, Group Policy, or – to particular delight – the Windows Registry. Microsoft warns administrators to back up the registry and understand how to restore it before making changes. After disabling the feature, administrators must restart the device and repair its secure channel using the Test-ComputerSecureChannel PowerShell command. As for the longer term, Microsoft said: "We plan to re
Watermarks that European law requires be added to AI-generated content to establish provenance may come at a cost. According to Lasso Security, AI model watermarking changes how AI agents handle tools and safety refusals. The altered behavior isn't necessarily worse but can be, particularly under adversarial prompt injection. With the implementation of the EU AI Act, providers of AI models must mark the output of their software with machine-readable code. Google DeepMind's SynthID-Text is one method for doing so, and has been adopted by Anthropic and by OpenAI. The benefit of this sort of digital labeling is that manipulative or deceptive AI-generated content can be more easily detected, even if it does have the potential to stigmatize the usage of AI. Anthropic's explanation of how it applies watermarks to Claude output involves intervening in the prediction that results in specific words. For example, if Claude were emitting the sentence "The weather today was cold and…" then it might favor one statistically likely candidate (e.g. "overcast") over an alternative (e.g "gray"). It may be possible to detect those additions. "Watermarking is designed for provenance, but SynthID-Text changes the process by which the model generates each next token," Lasso explained in a blog post provided to The Register. "At the model level, this can change safety behavior, including whether the model refuses a harmful request and whether that refusal holds under prompt injection." "Watermarking uses low-stakes choices like these – which occur many times over a piece of generated text – to leave a pattern in Claude’s responses," Lasso Security added. "That pattern is undetectable to the reader, but is detectable to anyone who has a key that encodes it." While a reader might not notice the word choice bias, AI agents can be subtly sensitive to vocabulary differences. Lasso found that this sort of digital content tagging can affect tool calling and refusal behavior. Watermarking, the co
An independent briefing for builders: the whole field read continuously, every story scored for relevance, and the noise left off the page.
300+ curated sources. Every story scored 1–10 for builder relevance by Claude's frontier model. The filler never makes it to the page.
GPUs, datacenters, power deals, and inference economics: the infrastructure layer that decides what every builder pays. Our signature coverage.
Every story is sourced. Every score is computed. We show our work and link to originals.
Every briefing closes with The Call: one falsifiable claim with a date on it. When we're wrong, we say so in print. Opinions are cheap; ours get scored.