Axios AI+

February 05, 2024
Hi, it's Ryan. First, the Apple Vision Pro takes Manhattan.... Today's AI+ is 1,278 words, a 5-minute read.
Situational awareness: Snap announced it was laying off 10% of its workforce, or more than 500 employees, per CNBC.
1 big thing: IBM leads AI patent race
Illustration: Shoshana Gordon/Axios
IBM — not Microsoft, Google or OpenAI — tops the list of firms with the most AI-related U.S. patent applications over the last five years, according to new research shared first with Ina.
Why it matters: While patents are just one indicator of research intensity, the increase in the total volume of AI-related patent applications indicates strong interest, especially among large tech companies — and IBM's lead could show that its AI push is more than just marketing.
According to an analysis by IFI Claims, IBM made 1,591 AI-related patent applications, followed by Google, Microsoft, Samsung, Intel and Adobe, Capital One and China's Baidu.
- More than one in five, or 22% of the AI-related patents, dealt with generative AI.
- GenAI patents are rising quickly, IFI Claims said, with granted applications growing at a compound annual rate of 16% over the past five years. while patent applications for the field have grown at a rate of 31%.
How it works: IFI Claims, which specializes in patent data, created a system to identify AI and generative AI-related patent filings.
- The company noted that with newer technologies, patent classifications take time to emerge.
- "The patent coding system often trails state-of-the-art technologies; as a result, they are assigned more general classifications," it said. "IFI tailored a query that used technologies currently in play for GenAI. As time goes on, the classifications will enlarge."
Between the lines: ChatGPT creator and AI pioneer OpenAI didn't even make the top 25 companies on the list. IFI Claims said it found just one patent application, which only recently became public.
- "This is surprising for an organization that is knowledge-based," IFI said.
- However, it noted that the company may have more recent filings that have not yet been made public.
- Or, IFI suggested, OpenAI could be more focused on protecting its intellectual property via trade secrets rather than patent filings, which are eventually made public.
- OpenAI declined to comment to Axios on its approach to patents.
Yes, but: Patents are just one measure of innovation. Not all companies focus on getting that type of protection for their innovations, and not all patent applications are granted.
- Also, there are no guarantees that a patented invention will ever become a feature or product.
The big picture: IBM had long been the overall leader when it comes to U.S. patents, but Samsung has topped the list the past two years for most U.S. patent awards.
- This year Qualcomm finished at No. 2, becoming the first American company in the last three decades to top IBM.
IBM has said it will take a more selective approach to patenting, focusing on key areas rather than patenting all its inventions.
- AI is one of the areas it has cited.
Of note: IBM's generative AI patents are split across image, speech, text and video, as are those from Google and Samsung.
- By contrast, Nvidia's patents mostly focus on images and video, while Apple's are centered on speech.
What to watch: Every one of these patents is for a human-made invention relating to AI. But one of the big emerging questions is how to handle inventions produced by AI itself as the technology advances.
- Generative AI is aiding the search for new materials and formulas for everything from batteries to medications.
- But the law hasn't yet settled on how to treat AI-generated inventions —and most forms of intellectual property protection require a human to have created the work.
2. Swift deepfakes show threat to non-celebs, too
Illustration: Natalie Peeples/Axios
AI has made it easy to generate realistic-looking fake porn and the targets of malicious deepfakes are finding they have little recourse, reports Axios' April Rubin.
Why it matters: When everyone with a computer can create a convincing and harmful image, anyone from high school teens to the world's biggest pop star could fall victim to these potentially damaging deepfakes.
- In the absence of federal legal protection — and a smattering of state laws on the issue — those affected can be left to deal with lasting consequences on their mental health and reputation.
- For ordinary people, self-protection against harassment "can be really challenging, if not impossible," Bernard Marr, a futurist and generative AI expert, tells Axios.
Driving the news: A bipartisan group of senators introduced a bill last week designed to hold people accountable for sharing "digital forgery."
- "Although the imagery may be fake, the harm to the victims from the distribution of sexually explicit deepfakes is very real," Judiciary Committee Chair Sen. Dick Durbin (D-Ill.) and ranking member Sen. Lindsey Graham (R-S.C.) said in a statement.
- If passed, the legislation would provide a civil remedy for victims.
Between the lines: Victims of harmful AI-generated images are limited in their legal options, Mary Anne Franks, the president of the Cyber Civil Rights Initiative, tells Axios.
- Even in states with relevant laws, the abuse is defined "in really, really specific terms that are hard for the average victim to meet," she says. Other hurdles include having to know who the original perpetrator was.
Catch up quick: Taylor Swift was the victim of text-to-image generated explicit photos, created using workarounds to Microsoft's anti-porn protections, 404 Media found.
- A different method to create explicit images requires as little as one photo of a person, as teenage girls at a New Jersey high school last year found out.
By the numbers: Fake nudes have increased more than 290% since 2018 on the top 10 websites that host them, Washington Post reported last year.
- 86% of people polled by the AI Policy Institute in January said they believe "it should be illegal" to use AI to create deepfake porn.
3. IBM researchers use AI voices to hijack calls
Illustration: Shoshana Gordon/Axios
IBM researchers have found a relatively easy way to hijack voice calls using generative AI tools, Sam Sabin reports.
Why it matters: Many financial institutions and other stewards of people's most sensitive data lean heavily on phone calls to verify identities.
What's happening: IBM's researchers detailed a new threat they're calling "audio-jacking" — where voice clones manipulate a large language model midway through an ongoing conversation.
How it works: A threat actor starts by installing malware on a victim's phone or compromising a wireless voice-calling service, connecting it to their own AI tools.
- An AI chatbot receives a simple prompt telling it how to respond when it hears certain phrases, such as "bank account."
- Once the chatbot hears the phrases, it replaces what was shared with a different phrase using a victim's cloned voice — for example to redirect money to a different account.
What they're saying: The technique "could also modify medical information, such as blood type and allergies in conversations; it could command an analyst to sell or buy a stock; it could instruct a pilot to reroute," Chenta Lee, chief architect of threat intelligence at IBM Security, wrote in the report.
Threat level: Cybersecurity experts have warned that generative AI is already making voice scams easier to believe.
Yes, but: In IBM's experiment, not all voice clones were convincing — sometimes there was a lag in the voice clone's response because it needed to access both the text-to-speech APIs and the chatbot telling it what to do.
4. Training data
- EU national governments have unanimously signed off on the EU AI Act — the last major hurdle to the law's implementation, which will start late this year. (Techcrunch)
- The deterioration of TikTok's feed puts the social network on "the precipice of a wasteland period." (The New York Times)
- Go inside a government social credit system — where life-changing decisions are made through AI-powered suspicions. (Wired)
- Sabbaticals for CEOs aren't a common thing, but Automattic's Matt Mullenweg just started his — making him the second CEO to take one at Wordpress' parent company.
5. + This
The most futuristic city — Chongqing, China, population 30 million.
Thanks to Scott Rosenberg and Megan Morrone for editing this newsletter.
Sign up for Axios AI+

Scoops on the AI revolution and transformative tech, from Ina Fried, Madison Mills, Ashley Gold and Maria Curi.


