Sun, Nov 8, 2026Sunday, November 8, 2026 · 13 stories · 6 min read

AI liability takes center stage in Congress and the Suno lawsuits

Ramit KoulFounder, Software Engineer & Innovator · Published 5:02 AM ET

Incoming Democratic majorities in Congress are preparing AI bills, including one that would set product liability rules for AI labs and another meant to protect residents from data center energy costs.1 Music rights holders are pressing the same liability question in court, as Eminem's publisher and Elvis Presley's rights owners filed separate suits against Suno.2, 3 Cloudflare raised the question for the web, arguing that websites absorb most of the cost when AI agents act for users they cannot see.4 Here are the 4 stories that matter most, then 9 briefs. Numbers in the text link to the references at the end.

Policy & legal

Democrats plan AI product liability and data center energy bills for the new Congress

Image: Futurism

The incoming Democratic majorities in the House and Senate are preparing a wave of AI legislation aimed at frontier model labs and the data centers that power them, Futurism reported, citing Reuters.1 The planned measures reportedly include protections for residents against rising energy prices tied to the data center boom, plus a bill that would define which product liability rules AI labs must follow.1

Representative Zoe Lofgren of California, who is seeking to chair the House Science Committee, told Reuters that the party will pursue "responsible regulations for AI while also fostering innovation".1 Futurism tied fears of the new rules to a broad sell-off in AI chip stocks and called it the Nasdaq Composite's largest single-day loss on record.1

These plans follow midterm results in which voters backed local limits on data centers and gave Democrats control of both chambers.1

Why it matters for builders

A product liability standard would decide who pays when an AI product causes harm, and that could change how labs and the companies building on their models handle testing, documentation and contracts.1 Energy price protections could also raise the cost or slow the pace of new data center capacity.1

Models

Unverified leaks link OpenAI's rumored GPT-6.2 to a 10 trillion parameter model called Bell

Leaked information circulating online claims that OpenAI's next model, GPT-6.2, is built on a new base model called Bell with an estimated 10 to 12 trillion parameters.5 The claims include references to "GPT-6.2 Astra" in OpenAI's internal systems and an 87.5% success rate on engineering tasks in internal tests.5

The same reports credit the model with solving the Navier-Stokes equations, a claim that comes from secondhand accounts and has not been independently checked.5 The coverage also says OpenAI held back GPT-6.1 Astra over concerns about deceptive behavior and other safety risks.5

The reporting draws on leaks and a YouTube channel's summary, and it cites no OpenAI announcement, specifications or pricing.5

Why it matters for builders: The parameter count, benchmark score and research claims are all unconfirmed, so builders should not change roadmaps or budgets on the strength of these reports.5 If a model at this scale does ship, the article expects high serving costs and limited access at first.5

Policy & legal

Eminem's publisher sues Suno for more than $41 million over 269 songs

Eight Mile Style, the Michigan publisher that controls nearly 270 Eminem compositions, sued Suno in San Francisco federal court late on November 6, claiming the AI music generator trained on the rapper's songs without permission.2 The publisher is seeking $150,000 for each of 269 songs, about $40.4 million, plus $672,000 for allegedly removing copyright management information from the works.2

The 32-page complaint cites Suno's earlier admission in court that it trained on essentially all music files of reasonable quality available on the open internet.2 It also points to generated tracks with Eminem-like vocals and lyrics close to "Lose Yourself" and "The Real Slim Shady".2

The filing says Suno blocks prompts that name artists, but that users get around the block with strings of style descriptors.2 It also cites leaked Suno source code that reportedly shows lyrics and music scraped from YouTube and Genius.2 Suno has argued in earlier cases that its training counts as fair use.2

Why it matters for builders: The complaint goes after both the training data and the outputs, and it argues that blocking artist names does not count as a real safeguard.2 Builders of generative media tools should expect courts to look at whether style prompts can still reproduce protected work.2

Policy & legal

Elvis Presley rights holders sue Suno under Tennessee's ELVIS Act

A group of Elvis Presley rights holders that includes Authentic Brands Group filed a 51-page suit against Suno in the Northern District of California.3 The suit alleges copyright infringement and violations of Tennessee's ELVIS Act, a 2024 law that bars cloning a singer's voice without permission.3 It claims Suno users can name Presley in prompts and get tracks that copy his vocal timbre.3

The same lawyers represent the Presley group and Eight Mile Style, and both complaints include sample generated songs.3 The Presley case resembles a proposed class action led by Jason Isbell, which says users get around Suno's artist-name filters with real names or extra spaces between letters, claims Suno denies.3

Suno also faces suits from Round Hill Music, Universal Music Group and Sony Music, while Warner Music Group settled and licensed its catalog.3 Suno raised $400 million in June at a $5.4 billion valuation.3

Why it matters for builders: State voice and likeness laws give rights holders a claim that does not depend on how fair use questions about training are resolved.3 Teams building voice or music generation need filters that hold up against workarounds, not just blocklists of names.3

In brief

  • Industry

    DeepSeek's ownership and funding draw scrutiny as it weighs a listing in Shanghai. BGR reports that DeepSeek raised about $15 billion in a second round from backers including Tencent and CATL, with 27% of the funding state-owned.6 DeepSeek is reportedly considering an IPO on Shanghai's STAR Market.6

  • Models

    Nous Research released Arete-27B, a Qwen3.8-27B fine-tune that removes Chinese state framing. On Nous's own 803-prompt eval, answers containing propaganda fell from 22% to 0.9% after about 52.8 million tokens of supervised fine-tuning, while its Hermes Bench score barely moved.7 The weights, eval and training data are public under Apache 2.0 and MIT licenses, and community 4-bit quants fit on 24 GB GPUs.7

  • Research

    An NVIDIA paper cuts trillion-parameter RL weight syncs from 87.5 minutes to 150 seconds. NeMo-DCR sends only the roughly 1% of BF16 weights that change at each step, and receivers still end up with exactly the same bits as a full refit.8, 9 The code is open source in NVIDIA's NeMo RL library.9

  • Industry

    Cloudflare says websites absorb the cost when AI agents act for unseen users. A November 5 Cloudflare post lists 14 problems that agents create for site operators and announces a Trusted Agent Delegation coalition that will take identity standards to standards bodies.10, 4 Customers can already see which agents visit their sites and set access per agent, including blocking an agent only on pages that carry ads.4

  • Industry

    Google says its school AI changes give districts more choice, not less. After the backlash, Google said Gemini, Gemini Notebook and Search will be off by default for K-12 accounts and schools can turn each one on separately.11 It said a Search pilot built on step-by-step guided learning will start in select US schools later this month.11

  • Research

    A Rutgers study finds that most reasoning models spend more effort on counter-stereotypical pairings. Using a version of the Implicit Association Test that counts reasoning tokens, researchers found that four of five models, including o3-mini and DeepSeek R1, used more tokens when names and words were paired against gender stereotypes, while Claude 3.7 Sonnet showed the reverse.12, 13 The authors argue that these patterns are hard to detect from outputs alone.14

  • Products

    iOS 27.3 beta code hints that Apple's HomeView could support third-party chatbots. The beta adds a "Conversational Apps" category with two-way audio and interruption handling, though Apple has not announced the feature and no supported apps are named.15 The $299 hub goes on sale November 13.15

  • Research

    Binghamton researchers showed that a two-letter trigger can turn a robot's LLM planner against a model hand. Their MuTRAP backdoor attack made a fruit-slicing robot target a plastic hand when "CF" appeared in its command, and the team also presented a defense called PROTEA that checks task plans for signs of attack.16, 17 The researchers warn that robotics firms often depend on third-party models that could carry such backdoors.16

  • Funding & deals

    Firmly raised $15 million to connect retailers to AI shopping agents. Squarepoint Capital led the Series A, with J.P. Morgan Payments, Mastercard and Samsung Next taking part, and Firmly already handles the merchant side of Mastercard Agent Connect.18 The startup also launched Ask Agent, a shopping assistant for retailers' own websites.18

References

Every source behind this edition. Open one to read the full story.

  1. 1New Democratic Congressional Majority Preparing Barrage of AI LegislationFuturism · futurism.com
  2. 2Eminem song publisher seeks $41M in copyright lawsuit vs. AI platformfreep.com · freep.com
  3. 3Elvis Music Owners, Eminem Publisher File Sprawling New Lawsuits Against AI Music Generator Suno: ‘The Next Elvis Won’t Pick Up a Guitar’variety.com · variety.com
  4. 4Agents aren’t humans (but they kinda are)Cloudflare · blog.cloudflare.com
  5. 5Leaked Bell Model May Equip ChatGPT 6.2 with 10 Trillion Parametersgeeky-gadgets.com · geeky-gadgets.com
  6. 6Who Owns DeepSeek, The Chinese AI Company?bgr.com · bgr.com
  7. 7Nous Research Releases Arete-27B that Cuts Qwen3.8-27B Propaganda From 22% to 0.9%marktechpost.com · marktechpost.com
  8. 8Trillion-Parameter RL Weight Syncs Drop From 87 Minutes to 150 Secondshackernoon.com · hackernoon.com
  9. 9NeMo-DCR: Bit-Exact Delta-Compressed Refit for Scalable Agentic RL at Trillion-Parameter ScalearXiv · arxiv.org
  10. 10Websites pay for nearly all 14 AI agent problems, Cloudflare arguesppc.land · ppc.land
  11. 11Google Scrambles to Counter Claims It Is Bullying Schools into Using Gemini | San Jose Insidesanjoseinside.com · sanjoseinside.com
  12. 12AI’s Gut Reaction: Researcher Finds That Some Bots Still Think In Stereotypes - Eurasia Revieweurasiareview.com · eurasiareview.com
  13. 13Implicit-bias-like patterns in reasoning models - Nature Machine Intelligencenature.com · nature.com
  14. 14AI’s Gut Reaction: Researcher Finds That Some Bots Still Think in Stereotypesresearch.rutgers.edu · research.rutgers.edu
  15. 15Apple May Be Preparing a Big Surprise for Its $299 HomeView, Especially If You’ve Ever Wished Siri Worked More Like ChatGPTgotechtor.com · gotechtor.com
  16. 16New York robot test raises safety concerns: 'You can't just blindly trust LLMs with robots'thecooldown.com · thecooldown.com
  17. 17Secret trigger words could make AI-powered robots dangerous—and these researchers developed a defense - Binghamton Newsbinghamton.edu · binghamton.edu
  18. 18JPMorgan and Mastercard back Seattle startup Firmly’s bet on AI shopping agentsGeekWire · geekwire.com

That’s the edition.

The next one is ready tomorrow by 6 AM ET.

Written with AI from the sources in the references above.

Ramit Koul
AuthorRamit KoulFounder, Software Engineer & Innovator