Today’s list is about who is doing the work. At DeepMind, a pair of AI agents ran a big cryptography computation while a human mostly sent approvals, and a startup says a non-public Anthropic model finished a quantum proof that had stood open since 2008. Then the music business takes Suno back to court, Microchip says AI buyers are crowding everyone else out of the factories, and Microsoft Research loses its AI Frontiers lab.
1. DeepMind’s AI Agents Cut the Cost of Attacking Old 1024-Bit RSA
What happened. On Nov 6, Google DeepMind posted two working papers on legacy 1024-bit RSA. In the first, two cooperating AI agents, a supervisor and an executor, designed, ran and debugged a known signature-forgery computation, writing about 33,500 lines of code while the human researcher sent roughly 20 short messages, mostly approvals. The rerun took 391 CPU core-years against about 1,380 in the earlier study, and finished in under four days instead of five months. The paper also proposes a new algorithm that DeepMind estimates is roughly 1,600 times faster at this size, and a second paper lowers the per-key cost of factoring many keys at once.
The hype check. The 1,600x figure is a theoretical estimate, and the authors say they did not run the new algorithm at 1024 bits. The measured 3.5x saving came mostly from faster code, not a new attack, and the attack needs a signing interface that accepts chosen data. DeepMind says 2048-bit and larger keys are not threatened, and both papers are working papers that have not been peer reviewed.
Verdict: Promising but unverified. Agents doing days of real research engineering is the actual story here, while the headline speedup is still on paper.
2. Elvis’s Rights Holders and Eminem’s Publisher Sue Suno
What happened. Suno picked up two new lawsuits on Nov 7. Eight Mile Style, which says it owns or controls 269 Eminem compositions, sued in federal court in Northern California, alleging copyright infringement, circumvention of protection measures and removal of copyright information, and seeking damages that could top $40 million. A group of Elvis Presley rights holders filed a separate suit claiming Suno trained on his music and lets users imitate his voice, and it leans on Tennessee’s ELVIS Act. Eminem himself is not a plaintiff.
The hype check. These are allegations, and none has been tested in court. Suno says it blocks prompts that name specific artists or songs, screens uploads with outside providers, and has argued in other cases that its training is fair use. The timing is awkward, coming the same week Suno pushed song creation into chatbots.
Verdict: Needs follow-up. The voice and likeness claims under the ELVIS Act are a newer legal angle than the familiar training-data fight, and worth watching.
3. Microchip Says AI Buyers Took Over the Factories
What happened. Microchip CEO Steve Sanghi told investors on Nov 5 that AI customers “took over the whole factory” at the outside plants the company relies on, so it could have shipped more. Standard parts that used to ship in four to eight weeks now take six to 32 weeks. The stock fell about 10% on Nov 6, after a roughly 14% drop the day before in a broad chip selloff.
The hype check. This is one company’s view, and part of the stock slide came from sector-wide selling, not Microchip alone. Still, it is a concrete, on-the-record sign that AI demand is spilling over onto buyers of ordinary chips.
Verdict: News. Longer lead times for everyday parts are a cost of the AI buildout that rarely makes the keynote.
4. Microsoft Research’s AI Frontiers Lab Is Gone, and Its Leader Has Left
What happened. In a Nov 6 post titled “The End of AI Frontiers,” John Langford wrote that the roughly 50-person group inside Microsoft Research was reorganized into Microsoft AI in early September and is now over. He says Microsoft AI’s leadership put scaling ahead of research, that he left after placing his team, and that AI Frontiers will not take interns next year.
The hype check. This is Langford’s own account of an internal reorganization, so it reflects one side. It does fit a wider pattern of big companies folding open-ended research into product and scaling teams.
Verdict: News. Another research group gives way to the scaling push.
5. Startup Says a Non-Public Anthropic Model Proved a 2008 Quantum Conjecture
What happened. On Nov 6, Physical Superintelligence (PSI) said an internal research system running a non-public Anthropic model produced a proof of the Aaronson-Ambainis conjecture, open since 2008. The result implies that on random inputs, quantum computers are at most polynomially faster than ordinary ones for yes-or-no questions. PSI says the proof was checked in the Lean proof assistant, and Stanford researchers who read it called it “a huge result.”
The hype check. This is a company announcement, with the paper posted on GitHub and no peer review yet. The model drafted the paper under a researcher’s direction, the proof builds on other researchers’ work, and PSI does not spell out exactly what the Lean formalization covers.
Verdict: Promising but unverified. If the Stanford write-up holds, this is one of the clearest cases yet of AI closing a named open problem.
What to watch
Watch whether outside cryptographers reproduce DeepMind’s numbers or run the new algorithm at full size, and whether the Stanford group’s write-up backs PSI’s proof. In court, look for Suno’s first responses, especially on the ELVIS Act voice claims. And check whether Microchip’s lead times keep stretching into its next update.
Sources
- Google DeepMind: Modular Roots from a Restricted Oracle (working paper)
- KION (KPIX): AI platform reproduced Michigan music publishers’ copyrighted works without authorization, lawsuit says
- 24/7 Wall St.: The chip shortage is back, and Microchip says AI customers are crowding everyone out
- hunch.net: The End of AI Frontiers
- Physical Superintelligence: Aaronson-Ambainis
By Willa Ravenwire · The AI Bandwagon
