Tag: ai

Anthropic Says ‘Evil’ Portrayals Of AI Were Responsible For Claude’s Blackmail Attempts

Fictional portrayals of artificial intelligence can have a real effect on AI models, according to Anthropic.

Last year, the company said that during pre-release tests involving a fictional company, Claude Opus 4 would often try to blackmail engineers to avoid being replaced by another system. Anthropic later published research suggesting that models from other companies had similar issues with “agentic misalignment.”

Apparently Anthropic has done more work around that behavior, claiming in a post on X, “We believe the original source of the behavior was internet text that portrays AI as evil and interested in self-preservation.”

The company went into more detail in a blog post stating that since Claude Haiku 4.5, Anthropic’s models “never engage in blackmail [during testing], where previous models would sometimes do so up to 96% of the time.”

What accounts for the difference? The company said it found that training on “documents about Claude’s constitution and fictional stories about AIs behaving admirably improve alignment.”

Related, Anthropic said that it found training to be more effective when it includes “the principles underlying aligned behavior” and not just “demonstrations of aligned behavior alone.”

“Doing both together appears to be the most effective strategy,” the company said.

Keep reading

ICE Agents Have List of 20 Million People on Their iPhones Thanks to Palantir

Immigration and Customs Enforcement’s (ICE) use of Palantir systems now means agency officials effectively have a list of 20 million people readily accessible on their iPhones, increasing the speed at which ICE can find houses to raid and people to arrest, according to comments made by a senior ICE official last week during a border security conference.

While ICE and the Department of Homeland Security (DHS) generally won’t answer questions from journalists about how the agency is using Palantir’s technology, senior officials were much more talkative during the Border Security Expo which took place in Phoenix, Arizona, last week. 404 Media spoke to four people who attended the conference. Here companies looking to sell their technology to ICE or other agencies gathered for two days of speeches, Q&As, and product pitches.

The officials’ comments may need to be taken with a pinch of salt, but still reflect ICE’s position that Palantir is allowing the agency to identify people to arrest and locations to raid faster. Although the Trump administration has attempted to step back from its mass deportation rhetoric and city wide raids, especially in the wake of killing multiple people, ICE continues to violently and wrongfully detain people. Data from April showed that 70.8 percent, or 42,722, of people held in ICE detention have no criminal conviction.

The four people who attended the Border Security Expo saw Matthew Elliston, assistant director of Law Enforcement Systems & Analysis at ICE, and other DHS officials speak.

At one point, Elliston made the comment about ICE agents having 20 million targets, or potential people to detain, on their iPhones. This list can lead ICE agents to an individual and a house; they can then see if another target might be next door. This target may be a lower priority, but ICE can now use that information to arrest more people.

At another point, Elliston said that Palantir’s technology has increased ICE’s rate of successfully locating a target from around 27 percent to just under 80 percent.

Keep reading

Meta launches WhatsApp ‘incognito’ mode to address privacy concerns for AI chats

Meta Platforms said Wednesday it’s rolling out an “incognito” mode for WhatsApp users to have private conversations with its AI chatbot, a move intended to ease privacy concerns about sensitive information that users share in chats.

The social media company said in a blog post that incognito chat mode provides a way to have private, temporary conversations with Meta AI, its artificial intelligence assistant that’s been available on WhatsApp for a few years.

Messages will be processed in a “secure environment” that even Meta can’t access, won’t be saved by default and will disappear when exiting a session, Meta said.

Generative AI systems have been dogged by privacy concerns because the large language models that underpin these systems are trained on vast troves of data, sometimes including personal information provided by users themselves in their conversations with AI chatbots.

Rival chatbot makers already have some privacy features. Google’s Gemini chatbot has the option to disable chat history and opt out of allowing one’s data to be used in training its AI models. ChatGPT has similar controls.

Meta says it’s rolling out incognito chats because users often ask chatbots sensitive questions or include private financial, personal, health or work data in their questions.

“We’re starting ask a lot of meaningful questions about our lives with AI systems, and it doesn’t always feel like you should have to share the information behind those questions with the companies that run those AI systems,” Will Cathcart, Meta’s head of WhatsApp, told reporters.

Incognito chat mode has safety features to prevent the chatbot from answering questions about harmful topics, Cathcart said.

It will “steer the user towards helpful information if it can and then refuse (to answer) and eventually even just stop interacting with the user completely,” Cathcart said.

Users will only be able to type in questions and get text responses; they won’t be able to upload or generate images. They’ll also have to confirm their age because Meta doesn’t allow users under 13 on its platforms.

Keep reading

Demi Moore at Cannes: ‘AI Is Here,’ ‘We Can Work With It,’ ‘You Fight It … Is a Battle We Will Lose’

Oscar-nominated actress Demi Moore declared “AI is here” at the Cannes International Film Festival while urging people to find ways to work with it.

Moore, who serves as one of the jury members, shared her thoughts on AI when Variety asked a question.

“Wow, that’s a big question. I think the reality is that to resist — I always feel that against-ness breeds against-ness. AI is here. And so to fight it is to fight something that is a battle that we will lose. So, to find ways in which we can work with it, I think, is a more valuable path to take,” she said.

“To your question of, are we doing enough to protect ourselves? I don’t know the answer to that. And so my inclination would be to say probably not,” she added.

Moore said people should not be afraid of AI, believing it will not replace human creativity.

“The truth is there really isn’t anything to fear because what it can never replace is what true art comes from, which is not the physical, it comes from the soul,” she said. “It comes from the spirit of each and every one of us sitting here, to each and every one of us who creates every day. And that they can never recreate through something that is technical.”

Keep reading

The Ascent of Mediocrity

Regular readers of Brownstone Journal have been graced with insight provided by many authors of diverse backgrounds and experiences. As a physician, I have found those authored by Dr. Joseph Varon to be exceptionally helpful in their insight into the state of medicine today. In particular, his essay, “When Physicians are Replaced With a Protocol,” struck a chord with me.

Perhaps it was my conscience, as I probably bear some responsibility for furthering this viewpoint, at least on a local level. You see, I once was a True Believer. It was plausible. It seemed so believable, so “scientific,” so simple. But it was a vicious hoax that, I am ashamed to say, took me in. Let me tell the story:

In the early 1990’s, medicine was under siege. The cost was rising at a steep rate, and some people saw an opportunity. Rather than looking at the rapid corporatization of healthcare and the proliferation of administrative costs, it was easy to shift the blame to the “providers. We were no longer “physicians,” but providers of a service. In truth, that is what we had become. The Health Equation had been shifted, whether intentionally or by accident. Just a few years before, physicians had directed patients to hospitals. Now, some bright businessperson, probably from The Wharton School or other such academic Ivory Tower, had seen the profit if the hospitals (or other corporate entities like insurance companies or A COMBINATION OF THE TWO) directed the patients to the physicians. It was like some financial martial arts reversal move…A perfect Sumi Gaeshi.

Keep reading

Secretary Doug Burgum Backs BYOP: Data Centers Must ‘Bring Your Own Power’ to Curtail Economic Impact

Data centers must “bring your own power” to curtail the economic impact, Interior Secretary Doug Burgum said during a Monday event with Breitbart News.

Addressing concerns about the economic impacts of artificial intelligence (AI) in terms of energy, Burgum acknowledged, “When you bring in AI, the rates go up because they use so much power.” However, he said he fully embraces BYOP – bring your own power.

Pointing to the Ratepayer Protection Pledge, Burgum said it is a fairly simple concept.

“If you’re one of the hyperscalers, and if you want to build out a data center, then you have to, like, you know, say BYOP. You’ve got to bring your own power,” the Interior Secretary said.

And in the instances they do not want to bring their own power, they have to at least be willing to allow themselves to be “curtailed.”

“You have to be willing during those peak moments, hours, days of the year, to say, yeah, you can shut my center off. And you say, well, that’s not possible. It’s not possible in some forms of AI, but if you’re just doing a training model, then you could say, yeah, we’re, you know, we’re willing, over a course of 365 days to be shut down 24 hours,” he explained. “We can be curtailed.”

“If somebody says yes to that, there’s a lot of places you go in the country where we actually have excess power in the spring and the fall – we generate, you know, hundreds of gigawatts more power than what we use,” Burgum said.

“All the grid of operators are trying to, you know, trying to make sure that on those peak cold days and peak warm days that we can still keep everything running and not have the, you know, the cascading collapse on this giant machine called the grid,” he said. “So there’s that.”

But ultimately, Burgum said these data centers have the option to “build your own power.”

Keep reading

Parents Sue OpenAI After Son Fatally Overdosed Following ChatGPT Drug Advice.

 WHAT HAPPENED: A Texas couple has filed a lawsuit against OpenAI, alleging that ChatGPTprovided their son with unsafe advice about drug use, leading to his fatal overdose in 2025. The family alleges that ChatGPT recommended a combination of kratom and Xanax, which proved lethal for their 19-year-old son, Sam Nelson.

 DETAIL: The lawsuit claims the teen repeatedly used ChatGPT for guidance on various substances and that the chatbot gradually shifted from refusing harmful requests to offering specific recommendations on drug intake and recovery. His parents, Leila Turner-Scott and Angus Scott, argue that the AI platform dispensed dangerous advice that it was unqualified to provide and failed to maintain adequate safety protections. The suit, filed in a California state court, seeks to hold OpenAI responsible for wrongful death and negligence, alleging their son would still be alive if stronger safeguards had been in place. OpenAI has not publicly responded in detail to the lawsuit, but it has previously stated that ChatGPT is designed to discourage harmful behavior and direct users to professional help. The case adds to a growing number of lawsuits accusing AI chatbots of contributing to dangerous or violent conduct, such as mass shootings and mental health crises, including recorded suicides.

 KEY QUOTE: “The chatbot is capable of stopping a conversation when it’s told to or when it’s programmed to… And they took away the programming that did that.” – Leila Turner-Scott, the victim’s mother.

 IMPACT: The case highlights growing concerns over the potential for AI platforms to provide unverified medical advice, raising questions about liability and the need for stricter safeguards. It also underscores broader debates about the role of AI and whether teenagers, the mentally ill, and other vulnerable people should have unsupervised access to it.

Keep reading

Target Hospitality Jumps As Data Center Boom Fuels Demand For Worker Camps

Target Hospitality shares jumped in premarket trading after the company announced a new contract to provide mobile housing solutions and related hospitality services for workers at data center construction projects.

The 48-month contract could generate upward of $750 million in revenue for Target Hospitality, which builds, owns, leases, and operates large temporary or semi-permanent “communities” for workers of major projects. The contract covers 3,370 beds.

Historically, Target Hospitality generated revenue from energy, natural resources, and government-related customers, but since the data center buildout boom, its temporary housing solution services have been in high demand.

The company said that since the start of the year, it has announced over $1.4 billion in multi-year contracts amid data center buildouts, representing more than 9,000 beds.

“These awards reinforce the scale, customer relevance and capital-efficient deployment capabilities of Target Hyper/Scale, while strengthening Target’s exposure to long-duration demand across AI-driven data center and related critical infrastructure development,” the company wrote in a press release.

CEO Brad Archer wrote in a statement that the company is “entering the next phase of our growth with strong momentum and increasing confidence in our long‑term strategy. Since February 2025, we have secured more than $2.0 billion of multi‑year contracts, including approximately $1.8 billion within our rapidly expanding WHS segment, meaningfully enhancing revenue visibility, supporting consistent cash flows and driving improved margin contributions. These wins position Target to further expand its presence across high-value end markets with long-term momentum.”

In premarket trading, Target Hospitality is up nearly 10%. On the year, the stock has surged 91%, as of Friday’s close.

Keep reading

AI Safety Institute Debuts with Big-Name Backers and a Censorship Agenda

Common Sense Media’s Youth AI Safety Institute arrived at the Danish Parliament this week and the guest list is stacked with people who think you can’t be trusted to speak freely online.

Hillary Clinton, Ursula von der Leyen, former Biden Surgeon General Vivek Murthy, Ofcom chief Melanie Dawes, and the head of an organization that wants to break end-to-end encryption are all gathering at Christiansborg Palace in Copenhagen to announce what they’d like to do next about AI and children.

The “next” part is where it gets concerning. The Youth AI Safety Institute, launched by Common Sense Media on May 5, says it will “complement efforts by regulators and policymakers to translate frameworks such as the EU AI Act, the Digital Services Act, and the UK Online Safety Act into practical protections for child-safe AI.”

Those three censorship laws represent the most aggressive government-directed speech suppression regimes currently operating in the Western world. The Institute isn’t questioning them. In fact, it wants to help implement them and push them further.

The summit, titled “Keeping Our Children and Families Safe in the AI Era,” is co-hosted by Common Sense Media, Save the Children Denmark, and Margrethe Vestager, who spent years as the European Commission’s executive vice president building the regulatory architecture that now lets EU officials order platforms to delete content.

More than 200 policymakers, tech executives, and civil society figures are expected. King Frederik X of Denmark is giving the opening address. The Duchess of Edinburgh will attend. Danish Prime Minister Mette Frederiksen is on the bill.

And so is Pinterest CEO Bill Ready, whose company helped pay for the Institute’s creation.

Keep reading

Georgia Data Center Secretly Used 29 Million Gallons of Water, Exposed by Residents’ Low Water Pressure

A major data center campus in Fayette County, Georgia, drew nearly 30 million gallons of water through unmetered connections before the issue surfaced due to complaints of low water pressure from nearby homeowners, county officials said.

The discovery, first reported by Politico, centers on the sprawling 615-acre QTS data center development located about 20 miles south of Atlanta. Quality Technology Services (QTS), owned by Blackstone, operates the site, which is one of the largest data center projects in the United States.

Fayette County investigators found that the campus had been pulling water through two connections the county was unaware of and had not properly billed. As a result, QTS was issued retroactive charges totaling $147,474. County officials estimated the unmetered usage covered roughly four months, while the company maintained the period was between nine and 15 months.

Vanessa Tigert, director of the Fayette County Water System, attributed the oversight to an administrative error that occurred during the county’s transition to smart meters.

“Fayette County is a suburb, it’s mostly residential, and we don’t have much commercial meters in our system anyway,” Tigert said. “And so we didn’t realize our connection point wasn’t working.”

A QTS spokesperson confirmed the company paid the retroactive charges immediately upon notification and said the unmetered usage stemmed from the county’s meter system upgrade.

No fines were issued. County officials emphasized they are maintaining a cooperative relationship with the developer.

The Fayetteville campus currently includes 13 buildings encompassing approximately 6.2 million square feet. It is part of a larger planned development that could eventually include up to 16 buildings.

The incident highlights growing tensions nationwide over the resource demands of data centers. Communities across the U.S. have become increasingly vocal about the strain these facilities place on local water supplies and electrical grids, leading to heightened opposition to new projects.

In a separate but related development, an Indianapolis City-County Council member’s home was shot at in April shortly after he supported rezoning for a data center project. The attack on Ron Gibson came days after a 6–2 vote approving the nearly 14-acre facility in the Martindale-Brightwood neighborhood.

Keep reading