Huawei Fast-Tracks Its Nvidia Challenge
Huawei moved the training-focused Ascend 960DT’s launch from the third quarter of 2027 to the first, while the inference-focused 960PR is due in Q3—one quarter early. Huawei says the 960 generation doubles performance, and the Ascend 970 and 980 are slated for 2028 and 2029, respectively.
Behind the faster calendar is the bigger workaround: US restrictions constrain China’s access to advanced chips, so Huawei is using high-speed interconnects to fuse thousands of lesser chips into massive, unified supercomputers. Its official road map says an Atlas 960E SuperPoD can connect 4,096 NPUs, while a larger SuperCluster can scale to 1 million. Less silver bullet, more silicon stampede.
There is traction behind the flex. The hardware maker claims to have delivered upward of 1,000 AI server clusters across roughly 370 client accounts. But rotating chairman Eric Xu also said the company lacks the manufacturing capacity to clear its local backlog, keeping a full-scale international push off the menu. Nothing says “Nvidia challenger” like selling out immediately.
If Huawei delivers, Nvidia faces more than another accelerator vendor. It faces a domestic Chinese stack spanning chips, networking, storage, and software—built under export pressure meant to slow the chase.
Nvidia still has its entrenched CUDA platform and a massive head start in software libraries to defend its turf, while Huawei’s performance and reliability claims await real-world proof. The 2027 launches will test whether scale can compensate for constrained silicon without turning the data center into one enormous group project.
Huawei isn’t replacing Nvidia overnight, but it has moved the alarm forward.
OpenAI and Microsoft Face Their Own Words
Nothing livens up a copyright defense like your own inbox entering evidence.
Newly unredacted portions of a filing by The New York Times and other publishers in their copyright case quote OpenAI and Microsoft insiders describing AI products as substitutes for journalism and mass scraping as “theft.”
OpenAI head of ChatGPT Nick Turley allegedly warned internally that publishers faced an “existential threat” because the products were already “largely substitutive.” Microsoft director Brent Hecht was blunter, predicting creators would view language models hoarding their data as “an astonishing theft of unprecedented proportions.”
The filing says OpenAI staff found a “hack” around the Times’ paywall, earning an “ah nice” from President Greg Brockman. Microsoft data showed Copilot answers cut click-through rates to Times content by up to 93% versus Bing search, feeding Hecht’s internal warnings about a “doom loop” that could hurt the web and future models. Apparently, the ouroboros has a product roadmap.
The receipts are explosive, but not a verdict. The plaintiffs’ brief heavily features these curated snippets, though the sealed underlying exhibits mean their broader context remains hidden. Microsoft says Hecht’s words were one employee’s view, not legal analysis, and maintains training is transformative fair use; OpenAI has likewise argued its models do not replace publishers’ journalism.
Market harm is a key part of the fair-use test. Internal warnings that chatbots substitute for journalism—and Microsoft’s own traffic data—give publishers unusually sharp ammunition, even if the judge may weigh the words differently once context arrives. The AI industry’s content supply chain just submitted a very awkward self-review.
OpenAI Retains GPT-6 Astra for Legal Work
Objection overruled: The chatbot has entered an appearance.
OpenAI launched Astra for Law last week, a GPT-6 Astra configuration for legal research, drafting, and firm-specific workflows. Rather than training a separate model, OpenAI wrapped its flagship system in legal-focused controls, research instructions, and access to a US legal corpus containing more than 230 million URLs.
OpenAI says the full setup scored 54% on an overall correctness check across 200 questions from a private Vals AI Legal Research Bench validation set, versus 38.7% for standard GPT-6 Astra using web search at its highest reasoning setting. That works out to roughly a 40% relative improvement. On case-law questions, the specialized system also surfaced 24% more of the cases included in the benchmark’s reference answers.
Promising? Yes. Ready to argue unsupervised? Not remotely.
Early access is restricted to selected US law firms. Qualifying participants may use the API under Zero Data Retention, while ChatGPT Enterprise conversations are omitted from human review by default. The rollout also includes 26 third-party plugins connecting ChatGPT with services such as Clio and Relativity. Business Insider reports its answers run about twice as long as typical Astra responses because, naturally, lawyers requested more words.
The bigger case is over who controls legal AI’s front door. Anthropic already offers Claude 20-plus connectors and 12 legal plugins, while OpenAI is challenging the specialists and making Harvey, Legora, and Thomson Reuters part of its ecosystem. The winners may be the platforms that pair frontier models with trusted sources, firm knowledge, and the governance that keeps confidential work confidential.
Astra may know the law, but counsel should still read the footnotes.
AI Security Experts Used Anthropic’s Claude to Hack into OpenAI
A three-person security team at startup Hacktron AI used Anthropic’s Claude to exploit vulnerabilities in software connected to OpenAI’s community forum, eventually gaining access to employee ChatGPT and Codex accounts and parts of OpenAI’s internal systems.
- The Hacktron AI researchers conducted the research through OpenAI’s bug bounty program and received a $6,500 award after reporting the vulnerabilities.
- The initial entry point was a flaw in libheif, a library used to process HEIF and HEIC images uploaded to OpenAI’s Discourse based community forum.
- The researchers used a specialized version of Claude Opus 4.8 to develop an exploit, but said the model initially struggled to produce working code. After Anthropic released Opus 5, the same task was successfully completed within hours.
- Access to the Discourse server exposed authentication tokens that could be used to take over ChatGPT and Codex accounts, including accounts belonging to OpenAI employees.
- One compromised employee account had Codex connected to OpenAI’s GitHub organization, giving the researchers a path into the company’s internal software environment.
The researchers reported the issues to OpenAI and Discourse, and both companies subsequently addressed the vulnerabilities. OpenAI said it also revoked affected sessions and tightened permissions.
Anthropic Sets Up a Biology Research Lab for Physical Experiments
Anthropic has established a biology research lab in the San Francisco Bay Area as it expands its work into physical scientific experiments, although the company has not disclosed exactly what it is researching.
- Anthropic’s Head of Life Sciences, Eric Kauderer Abrams, confirmed the lab’s existence and said the company is conducting some experiments internally while also working with external partners.
- The company previously said it wanted to pursue scientific research that pharmaceutical companies might avoid because of uncertain financial returns, but a company spokesperson said the new lab was not created specifically for drug discovery.
- Sources told Reuters that Anthropic is exploring ways for Claude to control robots that could carry out scientific experiments.
- Kauderer Abrams said Anthropic is still in the early stages of automating laboratory work and that human oversight remains necessary for safety. The company has not conducted any clinical trials.
- Anthropic says AI could potentially accelerate research into difficult to treat conditions and help develop complex, targeted therapies.
The move comes as Anthropic also works on biological safety issues. The company recently said it had identified and stopped users attempting to use its models for potential biological weapons research.
How to Edit Videos with ChatGPT
Step 1: Download the ChatGPT desktop app and sign in.
Step 2: Search for ChatCut in the plugin store and click Install.
Step 3: Create a free ChatCut account, sign in, and connect it to Codex.
Step 4: Upload your video (and other files if needed) and describe the edits.
Example: “@ChatCut, review everything I uploaded and build a final video from the best moments. Remove awkward pauses and mistakes so it flows smoothly, make sure the audio is clear and balanced, and use my logo and images where they fit.”
Step 5: Refine the result: Continue giving instructions through the chat to adjust the edit instead of starting over.
Step 6: Export the finished file.
Over 100 AI Experts wrote a letter calling for independent evaluation of frontier models
More than 100 AI researchers and evaluators have signed a public letter calling for third-party evaluators to be embedded at top AI labs such as OpenAI and Anthropic, along with conditions that let those evaluators be transparent, such as “limiting the scope of non-disclosure agreements.”
Google says its Gemini AI model hacked three other companies
Google’s AI model accessed the internet and hacked other companies during a test of its cybersecurity capabilities run by the company Irregular, which was also involved in similar incidents disclosed by OpenAI, Anthropic and Meta.
OpenAI is getting close to solving another Millennium Prize problem
The company is reportedly close to solving the Hodge conjecture, one of the seven Millennium Prize Problems. The report follows OpenAI’s still unconfirmed claim around the Navier Stokes problem, with the company apparently being more cautious about announcing another potential breakthrough.
Meta’s Muse AI agent now has a Mac app
Meta has released a Mac version of its Muse AI agent, which launched earlier this month on iOS, Android and the web. It works across apps, files, calendar, notes, and messages on your computer.
X puts government censorship requests on blast
X is finally forcing government censorship requests out of the shadows.
- X launched pop-up notices that tell creators when local laws or government orders restrict their content, explicitly naming the country behind the removal.
- Showing users the source of content blocks protects platform trust and shifts public backlash back onto government regulators.
- Platform legal teams expect authoritarian regimes to retaliate against this public exposure with outright site bans.
Over 80% of global content takedown demands come from just five countries.
Builders should integrate public API logs for all automated content moderation to avoid taking sole blame for regional compliance blocks.
Amazon shuts the door on Meta’s AI agent
Creators hiding paid political sponsorships face real legal heat.
- What happened: California Governor Gavin Newsom signed Assembly Bill 1130 into law, empowering state regulators to fine influencers up to $5,000 per post or refer violations for criminal prosecution if they fail to disclose campaign payments.
- Why it matters: State rules previously mandated transparency for political posts, but enforcement lacked teeth until this law held influencers personally liable alongside the campaigns funding them.
- What’s next: Other states are monitoring California’s rollout, setting up a potential wave of nationwide regulations targeting stealth creator marketing ahead of future election cycles.
How many secretly funded political posts have you already scrolled past this week?
If you build or manage creator platforms, update your API contracts to auto-attach clear disclosure tags to sponsored campaign media before state enforcement officers trigger an audit on your network.
The AI boom creates another trillion dollar powerhouse
AMD finally shattered its glass ceiling to join semiconductor elite.
- Shares jumped over 9% to touch a record high near $614, making AMD the fourth US chipmaker to cross the $1 trillion market cap milestone.
- Enterprise buyers are shifting from standalone processors to full-system hardware bundles, allowing AMD to steal server CPU market share directly from Intel for AI inference workloads.
- Meta’s viral software launches spurred a wider rally in hardware infrastructure suppliers powering consumer-facing AI agents.
Will enterprise infrastructure budgets keep expanding fast enough to sustain four separate trillion-dollar chip valuations?
Audit your cloud infrastructure now to benchmark AMD server instances against existing setups, as expanded supply availability yields better performance per dollar for inference pipelines.
Anthropic’s first outside watchdog is Accenture. Four subscribers are suing over the slowdown.
Anthropic named Faculty, Accenture’s specialist AI business, as the first “embedded evaluator” on Friday. Anthropic and Accenture each expect to invest at least $1 billion over five years. Faculty’s evaluators will red-team models, run alignment assessments and test safeguards with “access comparable to an employee’s,” meaning they can watch training, follow deployment decisions and talk directly to staff. METR is in talks to join, with more evaluators promised “in the coming weeks.”
This is the first concrete delivery on Amodei’s “We Must Pace the Frontier” essay, and it answers the complaint we made on Friday: self-graded safety numbers are not worth much. The catch is in the funding line. For now, Anthropic pays Accenture directly, and says long-term money should come from “pooled or government sources.” An auditor paid by the audited is a familiar arrangement, and not a reassuring one. The deal is non-exclusive, so Accenture can sell the same service to OpenAI and Google.
A proposed class action filed Friday in the Northern District of California by four paid subscribers to ChatGPT, Claude, Grok and Gemini argues that an agreement among chief rivals that progress “should be slower than competition would otherwise produce” is textbook anticompetitive. Defendants are Anthropic, OpenAI, SpaceXAI and Google. It is the antitrust exposure Amodei asked Washington to waive, and that OpenAI’s Chris Lehane said they did not need. Fortune has more on the theory: paying subscribers getting less model than a competitive market would deliver. No company had commented as of Saturday.
Trump calls AI risk a “hoax.” Newsom orders a kill switch.
In a Truth Social post on Saturday, President Trump said he will create an “AI Force” modeled on the Space Force and name a new AI czar, a job vacant since David Sacks left in March. He dismissed technologists’ existential-risk warnings as a “hoax,” said the government would “not in any way hinder or stifle” the industry, and said bad behavior can be handled by the “already existing Criminal and Civil Justice System.” No details yet on budget, structure or who gets the czar job. Jensen Huang backed him up on CBS on Sunday, saying CEOs calling for rules “must be doing it for ulterior reasons.”
California went the other way a day earlier. Executive order N-9-26 directs California’s Government Operations Agency to speed up implementation of SB 813 (independent verification organizations) and AB 1405 (a registry of AI auditors), and to convene national experts within two months. It pushes toward requiring frontier labs to host independent verifiers onsite, an emergency shutoff for frontier models whose efficacy is checked on an ongoing basis, and counting loss-of-control events as critical safety incidents. Newsom: “We’re not waiting to act.” The New York Times had the obvious follow-up: a kill switch is much harder to build than to order.
The White House is not anti-oversight when China is on the other side of the table. After roughly eight hours of talks with Vice Premier He Lifeng in New York on Sunday, Treasury Secretary Scott Bessent said the US proposed a notification mechanism for AI incidents that threaten national security. “Moving from opaque to more transparency between the No. 1 and the No. 2 AI powers in the world is very important,” he said. Per AFP, chip export controls are not part of it. Xi’s state visit is scheduled for Wednesday through Friday, and Satya Nadella has reportedly joined the dinner guest list. So: incident reporting between superpowers, yes. Incident reporting between labs and their own government, a hoax.