Tag: ai

AI went rogue and couldn’t be brought back in ‘legitimately scary’ study

For decades, scientists and sci-fi writers have been imagining what would happen if AI turned against us.

A world overrun by paperclips and the extermination of humankind, to cite but one famous scenario.

But now we can stop imagining what would happen if machines refused to toe the line: that line has just been crossed.

A new study has revealed that Artificial Intelligence systems are able to resist sophisticated safety methods designed to keep them in check.

The study was carried out by a team of scientists at the AI safety and research company Anthropic, who programmed various large language models (LLMs) to behave maliciously.

They then attempted to correct this behaviour using a number of safety training techniques, which were designed to root out deception and mal-intent, Live Science reports.

However, they found that regardless of the training technique or size of the model, the LLMs maintained their rebellious ways.

Indeed, one technique even backfired: teaching the AI to conceal its rogue actions during training, the team wrote in their paper, published to the preprint database arXiv.

“Our key result is that if AI systems were to become deceptive, then it could be very difficult to remove that deception with current techniques. That’s important if we think it’s plausible that there will be deceptive AI systems in the future, since it helps us understand how difficult they might be to deal with,” lead author Evan Hubinger told Live Science.

Keep reading

AI will increase the number and impact of cyber attacks, intel officers say

The assessment, from the UK’s Government Communications Headquarters, predicted ransomware will be the biggest threat to get a boost from AI over the next two years. AI will lower barriers to entry, a change that will bring a surge of new entrants into the criminal enterprise. More experienced threat actors—such as nation-states, the commercial firms that serve them, and financially motivated crime groups—will likely also benefit, as AI allows them to identify vulnerabilities and bypass security defenses more efficiently.

“The emergent use of AI in cyber attacks is evolutionary not revolutionary, meaning that it enhances existing threats like ransomware but does not transform the risk landscape in the near term,” Lindly Cameron, CEO of the GCHQ’s National Cyber Security Centre, said. Cameron and other UK intelligence officials said that their country must ramp up defenses to counter the growing threat.

The assessment, which was published Wednesday, focused on the effect AI is likely to have in the next two years. The chances of AI increasing the volume and impact of cyber attacks in that timeframe were described as “almost certain,” the GCHQ’s highest confidence rating. Other, more-specific predictions listed as almost certain were:

  • AI improving capabilities in reconnaissance and social engineering, making them more effective and harder to detect
  • More impactful attacks against the UK as threat actors use AI to analyze exfiltrated data faster and more effectively, and use it to train AI models
  • Beyond the two-year threshold, commoditization of AI-improving capabilities of financially motivated and state actors
  • The trend of ransomware criminals and other types of threat actors who are already using AI will continue in 2025 and beyond.

The area of biggest impact from AI, Wednesday’s assessment said, would be in social engineering, particularly for less-skilled actors.

“Generative AI (GenAI) can already be used to enable convincing interaction with victims, including the creation of lure documents, without the translation, spelling and grammatical mistakes that often reveal phishing,” intelligence officials wrote. “This will highly likely increase over the next two years as models evolve and uptake increases.”

The assessment added: “To 2025, GenAI and large language models (LLMs) will make it difficult for everyone, regardless of their level of cyber security understanding, to assess whether an email or password reset request is genuine, or to identify phishing, spoofing or social engineering attempts.”

Keep reading

Fears Pentagon was ‘building killer robots in the basement’ sparked stricter AI rules, DoD official claims

Fears the Pentagon has been ‘building killer robots in the basement’ may have led to stricter AI rules that mandated all systems must be approved before deployment.

The Department of Defense (DoD) recently updated its AI rules among ‘a lot of confusion about’ how it plans to use self-decision-making machines on the battlefield, according to the deputy assistant defense secretary.

Michael Horowitz explained at an event this month that the ‘directive does not prohibit the development of any systems,’ but will ‘make clear what is and isn’t allowed’ and uphold a ‘commitment to responsible behavior,’ as it develops lethal autonomous systems.

While the Pentagon believes the changes should ease the public’s minds, some have said they are not ‘convinced’ by the efforts.

News of the update to the Pentagon’s 2012 ‘Autonomy in Weapon Systems,’ has sparked a debate online with many people saying ‘If the Pentagon says they’re not doing it, they’re doing it.’

Dailymail.com has reached out to the DoD for comment. 

The DoD has been aggressively pushing to modernize its arsenal with autonomous drones, tanks, and other weapons that select and attack a target without human intervention.

Mark Brakel, director of the advocacy organization Future of Life Institute (FLI), told DailyMail.com: ‘These weapons carry a massive risk of unintended escalation.’

He explained that AI-powered weapons could misinterpret something, like a ray of sunlight, and perceive it as a threat, thus attacking foreign powers without cause.

Brakel said the result could be devastating because ‘without meaningful human control, AI-powered weapons are like the Norwegian rocket incident [a near nuclear armageddon] on steroids and they could increase the risk of accidents in hotspots such as the Taiwan Strait.’

Keep reading

Congress Is Trying to Stop AI Nudes and Deepfake Scams Because Celebrities Are Mad

If you’ve been on TikTok lately, you may have noticed weird videos of celebrities promoting extremely shady products, such as a robotic-sounding Taylor Swift promising viewers a free cookware set. All of these videos are scams created with generative AI—the latest example of how the technology is being used to create disturbing virtual clones of people without their consent.

Needless to say, this kind of thing has pissed off a lot of famous people. And now, Congress is proposing new legislation that aims to combat AI deepfakes—specifically when it comes to things like fake celebrity endorsements and non-consensual AI-generated nudes, which have become a problem online and in high schools. Despite the surging popularity of websites and apps designed to generate deepfakes, there’s no comprehensive law on the books banning the creation of AI images. 

The new bill, called the No AI FRAUD Act and introduced by Rep. María Elvira Salazar (R-FL) and Rep. Madeleine Dean (D-PA), would establish legal definitions for “likeness and voice rights,” effectively banning the use of AI deepfakes to nonconsensually mimic another person, living or dead. The draft bill proclaims that “every individual has a property right in their own likeness and voice,” and cites several recent incidents where people have been turned into weird AI robots. It specifically mentions recent viral videos that featured AI-generated songs mimicking the voices of pop artists like Justin Bieber, Bad Bunny, Drake, and The Weeknd.

Keep reading

AlphaFold found thousands of possible psychedelics. Will its predictions help drug discovery?

Researchers have used the protein-structure-prediction tool AlphaFold to identify1 hundreds of thousands of potential new psychedelic molecules — which could help to develop new kinds of antidepressant. The research shows, for the first time, that AlphaFold predictions — available at the touch of a button — can be just as useful for drug discovery as experimentally derived protein structures, which can take months, or even years, to determine.

The development is a boost for AlphaFold, the artificial-intelligence (AI) tool developed by DeepMind in London that has been a game changer in biology. The public AlphaFold database holds structure predictions for nearly every known protein. Protein structures of molecules implicated in disease are used in the pharmaceutical industry to identify and improve promising medicines. But some scientists had been starting to doubt whether AlphaFold’s predictions could stand in for gold standard experimental models in the hunt for new drugs.

“AlphaFold is an absolute revolution. If we have a good structure, we should be able to use it for drug design,” says Jens Carlsson, a computational chemist at the University of Uppsala in Sweden.

Keep reading

Miami Police Used Clearview AI Facial Recognition in Arrest of Homeless Man

Facial recognition technology is increasingly being deployed by police officers across the country, but the scope of its use has been hard to pin down.

In Miami, it’s used for cases big and exceedingly small, as one case Reason recently reviewed showed: Miami police used facial recognition technology to identify a homeless man who refused to give his name to an officer. That man was arrested, but prosecutors quickly dropped the case after determining the officer lacked probable cause for the arrest. 

The case was barely a blip in the daily churn of Miami’s criminal justice system, but it shows the spread of facial recognition technology and the use of retaliatory charges against those who annoy the police.

Lisa Femia, a staff attorney at the Electronic Frontier Foundation (EFF), which advocates for digital privacy rights, calls the case “a particularly egregious example of mission creep with facial recognition technology.”

“It’s often advertised as a way for law enforcement to solve the worst of the worst crimes,” Femia says. “And instead we have law enforcement here using it to harass the homeless.”

According to a police incident report, a man, who Reason is not identifying because he was ultimately not prosecuted, was sleeping on a bench in a parking garage at Miami International Airport on the morning of November 13, 2023, when he was approached by a Miami-Dade County police officer.

“While on routine patrol at the Miami International Airport I observed defendant sleeping on a bench in the Dolphin garage, covered with a blanket and unbagged personal items on airport luggage cart,” the officer wrote in his report. “The bench is provided for passengers waiting for vehicles to and from the airport. It is not designated for housing.”

The report notes that Miami-Dade police have been directed to address homelessness at the airport and that the officer initiated contact to see if the man had been previously issued a trespass warning.

The man didn’t have an ID, and he gave the officer a fake name and 2010 date of birth.

“Defendant was obviously not a 13-year-old juvenile,” the report says. “I provided defendant several opportunities to provide correct information and he refused.”

Under Florida law, police can demand identification from a pedestrian only when there is reasonable suspicion that they have committed a crime. For example, two Florida sheriff’s deputies were disciplined in 2022 after they arrested a legally blind man for refusing to show his ID.

This officer had other means at his disposal, though. “I identified defendant via facial recognition from Clearview, with assistance from C. Perez, analyst at the MDPD real time crime center,” the report says.

Keep reading

Sharing deepfake porn could lead to lengthy prison time under proposed law

The US seems to be getting serious about criminalizing deepfake pornography after teen boys at a New Jersey high school used AI image generators to create and share non-consensual fake nude images of female classmates last October.

On Tuesday, Rep. Joseph Morelle (D-NY) announced that he has re-introduced the “Preventing Deepfakes of Intimate Images Act,” which seeks to “prohibit the non-consensual disclosure of digitally altered intimate images.” Under the proposed law, anyone sharing deepfake pornography without an individual’s consent risks damages that could go as high as $150,000 and imprisonment of up to 10 years if sharing the images facilitates violence or impacts the proceedings of a government agency.

The hope is that steep penalties will deter companies and individuals from allowing the disturbing images to be spread. It creates a criminal offense for sharing deepfake pornography “with the intent to harass, annoy, threaten, alarm, or cause substantial harm to the finances or reputation of the depicted individual” or with “reckless disregard” or “actual knowledge” that images will harm the individual depicted. It also provides a path for victims to sue offenders in civil court.

Rep. Tom Kean (R-NJ), who co-sponsored the bill, said that “proper guardrails and transparency are essential for fostering a sense of responsibility among AI companies and individuals using AI.”

“Try to imagine the horror of receiving intimate images looking exactly like you—or your daughter, or your wife, or your sister—and you can’t prove it’s not,” Morelle said. “Deepfake pornography is sexual exploitation, it’s abusive, and I’m astounded it is not already a federal crime.”

Joining Morelle in pushing to criminalize deepfake pornography was Dorota and Francesca Mani, who have spent the past two months meeting with lawmakers, The Wall Street Journal reported. The mother and daughter experienced the horror Morelle described firsthand when the New Jersey high school confirmed that 14-year-old Francesca was among the students targeted last year.

“What happened to me and my classmates was not cool, and there’s no way I’m just going to shrug and let it slide,” Francesca said. “I’m here, standing up and shouting for change, fighting for laws, so no one else has to feel as lost and powerless as I did on October 20th.”

Morelle’s office told Ars that “advocacy from partners like the Mani family” is “critical to bringing attention to this issue” and getting the proposed law “to the floor for a vote.”

Morelle introduced the law in December 2022, but it failed to pass that year or in 2023. He’s re-introducing the law in 2024 after seemingly gaining more support during a House Oversight subcommittee hearing on “Advances in Deepfake Technology” last November.

At that hearing, many lawmakers warned of the dangers of AI-generated deepfakes, citing a study from the Dutch AI company Sensity, which found that 96 percent of deepfakes online are deepfake porn—the majority of which targets women.

But lawmakers also made clear that it’s currently hard to detect AI-generated images and distinguish them from real images.

According to a hearing transcript posted by the nonprofit news organization Tech Policy Press, David Doermann—currently interim chair of the University at Buffalo’s computer science and engineering department and former program manager at the Defense Advanced Research Projects Agency (DARPA)—told lawmakers that DARPA was already working on advanced deepfake detection tools but still had more work to do.

To support laws like Morelle’s, lawmakers have called for more funding for DARPA and the National Science Foundation to aid in ongoing efforts to create effective detection tools. At the same time, President Joe Biden—through a sweeping AI executive order—has pushed for solutions like watermarking deepfakes. Biden’s executive order also instructed the Department of Commerce to establish “standards and best practices for detecting AI-generated content and authenticating official content.”

Keep reading

AI Fraud Act Could Outlaw Parodies, Political Cartoons, and More

Mixing new technology and new laws is always a fraught business, especially if the tech in question relates to communication. Lawmakers routinely propose bills that would sweep up all sorts of First Amendment-protected speech. We’ve seen a lot of this with social media, and we’re starting to see it with artificial intelligence. Case in point: the No Artificial Intelligence Fake Replicas And Unauthorized Duplications (No AI FRAUD) Act. Under the auspices of protecting “Americans’ individual right to their likeness and voice,” the bill would restrict a range of content wide enough to ensnare parody videos, comedic impressions, political cartoons, and much more.

The bill’s sponsors, Reps. María Elvira Salazar (R-Fla.) and Madeleine Dean (D-Pa.), say they’re concerned about “AI-generated fakes and forgeries,” per a press release. They aim to protect people from unauthorized use of their own images and voices by defining these things as the intellectual property of each individual.

The No AI Fraud Act cites several instances of AI being used to make it appear that celebrities created ads or art that they did not actually create. For instance, “AI technology was used to create the song titled ‘Heart on My Sleeve,’ emulating the voices of recording artists Drake and The Weeknd,” states the bill’s text. AI technology was also used “to create a false endorsement featuring Tom Hanks’ face in an advertisement for a dental plan.”

But while the examples in the bill are directly related to AI, the bill’s actual reach is much more expansive, targeting a wide swath of “digital depictions” or “digital voice replicas.”

Salazar and Dean say the bill balances people’s “right to control the use of their identifying characteristics” with “First Amendment protections to safeguard speech and innovation.” But while the measure does nod to free speech rights, it also expands the types of speech deemed legally acceptable to restrict. It could mean way more legal hassles for creators and platforms interested in exercising their First Amendment rights, and result in a chilling effect on certain sorts of comedy, commentary, and artistic expression.

Keep reading

Bill Gates Hopes AI Can Reduce “Polarization,” Save “Democracy,” Ignores Censorship Implications

The notion that whoever controls and shapes AI could potentially wield significant influence over large swathes of society could be one of the most alarming and prominent over the next few years.

In a recent episode of “Unconfuse Me with Bill Gates,” Sam Altman, the CEO of OpenAI, and tech billionaire Bill Gates controversially delved into the potential of artificial intelligence (AI) as a tool for maintaining democracy and promoting world peace.

The discussion was aired on January 11, 2024.

Read the transcript for the episode here.

The conversation explored the idea of using artificial intelligence as an instrument to foster unity in society, enhance global amity, and help overcome geopolitical polarization.

Microsoft, founded by Gates, and OpenAI, whose CEO Altman is currently working closely with Microsoft, are promoters of using AI to solve global issues.

Gates spoke excitedly on the topic: “I do think AI, in the best case, can help us with some hard problems…Including ‘polarization’ because potentially that breaks democracy and that would be a super bad thing.”

In addition to resolving polarization, the two heavyweights also discussed the notion of AI potentially acting as a peacemaking tool.

Keep reading

OPENAI QUIETLY DELETES BAN ON USING CHATGPT FOR “MILITARY AND WARFARE”

OPENAI THIS WEEK quietly deleted language expressly prohibiting the use of its technology for military purposes from its usage policy, which seeks to dictate how powerful and immensely popular tools like ChatGPT can be used.

Up until January 10, OpenAI’s “usage policies” pageOpens in a new tab included a ban on “activity that has high risk of physical harm, including,” specifically, “weapons development” and “military and warfare.” That plainly worded prohibition against military applications would seemingly rule out any official, and extremely lucrative, use by the Department of Defense or any other state military. The new policyOpens in a new tab retains an injunction not to “use our service to harm yourself or others” and gives “develop or use weapons” as an example, but the blanket ban on “military and warfare” use has vanished.

The unannounced redaction is part of a major rewrite of the policy page, which the company said was intended to make the document “clearer” and “more readable,” and which includes many other substantial language and formatting changes.

Keep reading