Scroll has released an emergency fix for a bug that was found in a contract called ZkEvmVerifierPostFeynman. The contract checks the proofs that let Scroll post new states to Ethereum (ETH). Scroll’s post-mortem on its governance forum says the contract built its call to OpenVM’s Halo2 verifier by hand and used the first four bytes of a user’s proof as the call’s label. OpenVM’s verifier has a function called verify with the label 0x24270d54. So, if a proof began with those four bytes, the call went to verify instead of the normal path.

Scroll says the new fix denied attackers control

The platform said the function took its values from places the attacker could control. An attacker could use this to pass off an old, real proof as proof for a new input. The result would be a fake state root and a fake withdrawal root locked in on Ethereum. Older versions were safe because their verifiers had no function selector, so every call went down the normal path. The fixed contract now sends every call through one set entry point. This way, the contract sets the key details instead of the proof.

Scroll thanked the ethical hacker ZHOUCHUNGONG, who reported the bug on Nov. 3, one week before it published the fix. Scroll says no one got a chance to exploit the flaw. An attack using the flawed contract would have needed either the permissioned prover role or “enforced batch mode.” That mode only opens after seven days pass with no finalization, and that did not happen. Scroll rechecked all 748 finalizations made while the flawed contract was still live, and it also replayed 233 past finalizations on a copy of Ethereum mainnet, and they all ended with the same roots.

In the past, outsiders helped oversee Scroll’s emergency fixes. For instance, in May 2025, Scroll pushed a fix to its FRI verifier and a verifier swap in February 2026 that were both approved by the Scroll Security Council. The council had 12 members, and 9 of them had to agree before any action. The council was removed on June 1, 2026. A 3-of-4 admin multisig run by the Scroll team replaced it. Scroll lost its Stage 1 decentralization rating and is now listed at Stage 0. Notably, the multisig can upgrade core contracts, including the bridge, with no delay.

Scroll first proposed that the council be removed in April, saying that it cost more than it was worth. Some people on the forum disagreed and asked why Scroll did not just keep a smaller council. They also asked who would approve the network’s actions going forward. A proposal was posted to the Scroll DAO forum on Nov. 4, 2026, that would make the chain a private, invite-only network built around AI if it ever gets approved. The proposal also states that the SCR token would move back to the Ethereum mainnet, where holders would use it to vote and stake.

Notably, by April this year, Scroll’s total value locked (TVL) had shrunk from around $600 million in 2024 to roughly $24.3 million. TVL is down to about $6.6 million as of this report. The SCR token has also slid to about $0.018, recording a nearly 99% decline from its 2024 high.

The post Scroll fixes bugs that may have been used to fake withdrawals first appeared on Coinfea.

The Wall Street Journal said Wednesday that the board was looking into firing Jensen because of his connection with a woman who may be working for Chinese intelligence. The directors were set to meet as early as Wednesday to decide on his fate. But the story did not show that a vote had already been made or that Jensen had been fired. Such a potential transition could not have come at a worse time for Nvidia. There are questions about the ability to sustain the massive sums of money being put into artificial intelligence infrastructure.

Nvidia directors convene meeting to decide Huang’s fate

The story in The Wall Street Journal came on the heels of new allegations made in The New York Times about Kimberly Powell, Nvidia’s vice president of healthcare. As Cryptopolitan reported, Jensen and Kimberly had a romantic relationship despite both being married to other people. The newspaper also reported that the pair were seen behaving affectionately at a hotel in San Jose, California, during March.

Reportedly, their interactions were recorded, although we have not viewed the footage. Kimberly has a very important position at Nvidia. She is in charge of its global health care operations, which encompass the computing hardware and software used in medical imaging, pharmaceutical research, biological sciences and health data analysis. This includes the responsibilities in industries that Nvidia hopes to serve through AI technology.

The Nvidia stock price is down by 7% this week, on track to suffer from its sixth successive weekly drop. Meanwhile, investors have become more worried about the speed at which the world’s largest technology companies are burning cash on AI infrastructure and whether they can recoup those investments in future revenue. Political developments have been another complication. Last week’s U.S. midterm elections apparently gave Democrats commanding majorities in both chambers of Congress.

Voters also passed dozens of local measures that could make building data centers more difficult, raising questions about expansion plans across the industry. This is important for Nvidia, whose processors run the computers that operate large AI models. These are the chips companies use to train ever more demanding software and run workloads once models are deployed. Jensen is so closely identified with the business that executives, customers and competitors refer to him by his first name. His potential exit would raise an immediate question of leadership at one of the largest suppliers bankrolling the world’s AI investment.

The post Nvidia set to fire Hang as scandal continues to escalate first appeared on Coinfea.

In the launch post, Google DeepMind SVP Koray Kavukcuoglu wrote that Carbon topped CursorBench 4.0, Perplexity’s WANDR, and Zapier’s AutomationBench. Google says it costs less than half what the next closest outside model costs per task. Google built Carbon for long, multi-step jobs. Complex coding and debugging, open-ended scientific research, and cybersecurity are its targets. Separately, Google cooperated with Artificial Analysis to build the Multimodal Knowledge Agent benchmark. It assesses how models reason over dense files such as charts, blueprints, and tables.

Google says Carbon was created at under half the cost of other agents

Carbon was in second place. Google says it operated at an average cost per task more than six times lower than the model that beat it. Carbon matched the best model tested on Hebbia’s E2E Financial Agentic evaluation and was first of 23 frontier model configurations on the Balyasny Asset Management General Financial Intelligence benchmark. Carbon goes deeper and runs longer verification loops before it answers, which means it uses more tokens than its sibling model, Gemini 4 Argon.

Google is retaining Argon, the model it debuted last month. The company said that Argon is still the recommended choice for most software engineering and knowledge work. Carbon is left for things that stretch the limits of a model’s intelligence. It was released in two steps, with cybersecurity defenders getting early access via the Fairwind Program two weeks ago. Both models have the same introductory API price of $2 per million input tokens and $10 per million output tokens. Google said Carbon’s rates will leap to $4 and $20 after the introductory period ends November 30.

Cryptopolitan reported that Argon’s price opened below Anthropic’s Claude Opus 5.5 ($4 and $20) and OpenAI’s GPT-6 Astra ($10 and $50). Paid-tier developers are able to build with Carbon in Google AI Studio and Android Studio. Enterprises can obtain it in Gemini for business. Google AI Ultra subscribers pay $100 per month and get access to Carbon in the Gemini app, AI Mode in Search, and Antigravity. Google is previewing /swarm inside Antigravity, which converts one prompt into a team of coordinated agents.

Developers are able to pin Carbon to the roles requiring the greatest reasoning, such as architecture and auditing, while the lighter models manage the parallel work. Four weeks ago, Google said it was taking applications for 10,000 scientists to use Argon in Antigravity for free for six months. Thousands onboarded, and current and future participants get Carbon too. Early testers put Carbon to work in security and biology. Wiz’s team detected 0-day flaws in two major SaaS providers in under two hours, Google said.

The post Google boasts about Carbon’s prowess against rival agents first appeared on Coinfea.

The loss is a major change for the Tokyo-based company. SoftBank made a profit of 2.5 trillion yen in the same period a year earlier, the AP reported. However, its revenue continued to grow, rising 16% to 2.2 trillion yen from 1.9 trillion yen a year earlier. SoftBank’s profits can change quickly because they depend on how the company’s investments perform. The company owns stakes in OpenAI and chipmaker Arm Holdings, among others, via its Vision Funds.

SoftBank records loss after profit early in the year

The funds invest in tech startups and large technology companies. As is well known, the values of these companies are prone to sharp rises or dips. SoftBank founder Masayoshi Son built the company around internet technology before expanding into AI, telecommunications, robotics and online shopping. He is also close friends with Nvidia CEO Jensen Huang.

Son sold his Nvidia shares before the reports linking Huang to an alleged affair with a Chinese government agent emerged. The matter is reportedly under investigation by Nvidia and US authorities. U.S. President Donald Trump said the FBI was looking into another matter and expressed confidence that Huang would be fine during a White House briefing on Monday, according to the AP. The reports have also raised questions about whether China has gained access to Nvidia’s AI chips.

SoftBank’s shares have lost almost a third of their value in the past year. However, they rose nearly 14% in Tokyo on Tuesday before the earnings report. SoftBank’s results show how much it relies on a few major tech investments. Changes in their value can have a big impact on its profits, making its financial results harder to predict.

The post SoftBank reveals $7 billion quarterly loss as AI bets sour first appeared on Coinfea.

Google raised this concern at a hearing on Monday that also drew objections from Meta, TikTok and Snap. Rachel Lord, who heads policy for Google-owned YouTube in Australia and New Zealand, told the hearing that the problem lies in one part of the bill. This part requires age checks on any service where children could come across pornography or fetish content. Lord claims the wording is too broad and could affect platforms that were never designed to carry such material, Reuters reported. “This could require us to age-assure and block access to everyday communication services like messaging, email and private file storage,” she told the committee.

Google says rules could also affect access to Google Drive

During the hearing, Lord said the rules could also affect Google Drive. Liberal senator Sarah Henderson asked if Gmail and Drive might have to check users’ ages and scan their private emails, messages and files. Lord answered affirmatively, according to The Sydney Morning Herald. Google believes the rules should only cover services mainly used to share pornography. Lord also said the bill does not do enough to protect privacy.

She said an earlier draft released in September allowed companies to consider privacy, but that part was removed from the version now before the Senate. “We’re a bit confused about why it’s been removed,” she said, per the Herald. “There are no protections for Australian privacy when we’re talking about what are effectively private communications and private file storage.” The bill says companies will be judged on how well they protect users and whether their actions match the risks involved.

However, it does not mention privacy and says cost cannot be used as a reason for their decisions. Henderson called the bill “deeply flawed” and said it could force companies like Google to monitor people’s emails on a large scale. Communications Minister Anika Wells introduced the Digital Duty of Care bill on October 15. It would require online companies to prevent harm before it ever happens. Companies that fail to follow the rules could face fines of up to A$109 million (about $75 million). The same fine would also apply if they break the age-check rules.

Other tech companies have objected

The bill passed the House of Representatives in late October and now needs Senate approval. The Labor government is expected to seek support from the Greens, who back the duty of care but want higher fines for companies that break the rules. A committee is expected to report on the bill by November 20, and Wells wants it passed before the end of 2026. If approved, the new rules will take effect 12 months later. A spokesperson for Wells dismissed concerns that the bill could block children from using Gmail. “The government does not intend to stop children from using email or the internet,” the spokesperson said.

They also explained that the goal was to keep children away from pornography, not block basic online services. Wells has compared the rules to safety standards for food, cars and children’s toys. She has also added that the bill is not a form of censorship. Google was not the only company with concerns. Representatives from Meta, TikTok and Snap told the committee that the bill gives regulators too much power to decide what content or behavior is harmful. They also warned that platforms could end up removing more content than necessary just to avoid fines.

The bill would also replace some of the rules that these companies currently follow. Greens senator Sarah Hanson-Young took the opposite line, telling TikTok that “we actually need some proper rules in place because your community guidelines clearly aren’t cutting it,” the Herald reported. Australia has already introduced strict rules on children’s access to social media. Since December 2025, children under 16 have been banned from having accounts on major social media platforms, including YouTube.

The post Google says Australian safety bill could cut teens’ access to Gmail first appeared on Coinfea.

The deal was announced at Adobe MAX, Adobe’s creativity conference. It involves Firefly Foundry, a platform that allows companies to build AI models using their own content. Disney’s models will only use material from its own library, and any content they create will remain private and belong to Disney. A key part of the deal is that Disney’s AI models will use its own content for training. Adobe said this will help them create images and other content based on Disney’s characters, settings and visual styles that companies can safely use, FT.com reported.

Adobe and Disney to build content based on Disney’s characters

The toolkit has three main features. It can create images in Disney’s style and turn rough sketches into finished images. It can also convert simple designs into 3D models that designers can edit and view from different angles. The deal will affect several parts of Disney’s business. Its theme parks, resorts and cruise ships will use the AI models to speed up projects. Walt Disney Studios will use the tools for early design work through its StudioLAB team.

ESPN will also use the technology to reach sports fans and make more of its archived content available. Adobe will become an official AI partner to Disney. The deal gives Disney early access to Adobe Firefly, Premiere and Photoshop, as well as support from Adobe’s engineers. Neither company has revealed the exact value of the deal. The two companies have already tested the tools. They launched two AI models on June 16, 2026, through Walt Disney Imagineering, the team that designs Disney’s parks and attractions.

One of these tools helped to turn sketches into images. The test project cut work that usually takes months to just a few days. Kyle Laughlin, Disney’s senior vice president of technology, engineering and R&D, told Reuters that the tools have helped speed up creative work. Tasks that once took months can now be completed in hours or days, he stated. Laughlin added that the tools are not meant to replace Imagineering staff.

The deal also shows how big companies want to use AI. Many popular image generators were trained on content taken from the internet, raising questions about ownership of the material and the images created. Disney plans to avoid these issues by training its AI models only on content it owns. For Adobe, working with Disney shows that Firefly Foundry can help companies build AI models using their own content. For Disney, the main benefit is getting work done faster while keeping its familiar style.

There are still questions about how much Disney can do with the tools. Since the AI models are trained on Disney’s own content, they can only use material the company has already created. Using them across its studios, resorts and ESPN could also prove to be a bigger challenge. The deal comes as Disney considers major cost cuts. Bloomberg News reported that the company was looking at cutting thousands of jobs and reducing its yearly spending by 20%.

The post Adobe inks deal with Disney to build custom AI models first appeared on Coinfea.

If you have access, you’ll get a text message from a contact called “Alexa Shopping.” Once it’s set up, you can just text it like a regular friend. Amazon gave a few examples: you can start your weekly grocery run, reorder detergent you bought a month ago, or send a photo of an outfit to find similar clothes on the site. The agent can add things to your cart and help you make purchases. It can also monitor price drops and allow pricing history checks up to a year. The shopping agent also manages shopping lists, tracks packages, and can even find items outside of Amazon using a feature called Shop Direct.

Amazon breaks down new feature to users

Amazon said the agent learns preferences over time. Tell it to show only organic or vegan options, and it remembers for later. Amazon’s vice president of Conversational Shopping, Rajiv Mehta, stated that customers can now text Alexa with what they want to buy and let the assistant find it for them. Right now, Amazon is only opening the text feature to a small group of invited US customers. Millions more are expected to get access over the next few weeks.

For customers who have not been invited, there is the option of a waitlist on their sign-up page. You can start a shopping request in a text message and seamlessly pick it up later on the Amazon app, website, or an Alexa device—or the other way around. Yahoo Finance reported that Amazon hasn’t said how much money it expects to make from the new feature. The launch comes as Amazon tries to stop other shopping agents from using its platform.

In September, it blocked Meta’s Muse shopping tool. The company claimed the tool had accessed Amazon without permission and stored customer login details, according to TechCrunch. Amazon also took legal action against Perplexity’s Comet shopping agent and won a court order to block it in March. However, a US appeals court overturned the ban in August, and the case is still ongoing. Other companies are racing for the same market.

DoorDash recently shared that outside AI agents have only placed a few hundred orders on its app, so it’s focusing on its own text assistant instead. Other tools like Meta’s Muse and Google’s Universal Cart are jumping in to let AI shop and buy items for customers. By keeping customers on its own assistant, Amazon can maintain control over how they discover and buy products, rather than letting other AI tools take over that process.

The post Amazon shoppers can now purchase goods through texts with Alexa first appeared on Coinfea.

Microsoft has been using less of the rented computing power in recent months and has decided not to renew the deal, Bloomberg reported, citing people familiar with the matter. Neither company has shared how much the deal was worth. According to the reports, the rented capacity no longer met Microsoft’s needs. Oracle’s hardware used older Nvidia (NVDA) chips, making it less suitable for the AI tasks Microsoft is currently interested in running.

Microsoft says the need for computing power has reduced

An Oracle spokesperson said demand for its AI services remains strong, including for older GPUs used to train and run AI models. Oracle had announced in September that it was renewing contracts for four-year-old GPUs. These GPUs were going for prices 20% higher than initially sold. This raised a question of whether Microsoft was leaving because of a falling demand for Oracle’s cloud services or simply because it wants newer AI chips.

The deal started in 2023, when Microsoft rented computing power from Oracle to support a new version of Bing with AI features. At the time, there was not enough computing power to meet the growing demand from AI projects. The deal was seen as unusual back then. Microsoft, one of the world’s biggest cloud providers, was relying on a competitor for extra computing power.

However, it was a temporary fix that allowed Microsoft to keep its AI services running for business customers while building more data centers of its own. Once they were ready, the company planned to move the work to its Azure cloud. The news came as tech stocks were already falling. On Thursday, the Nasdaq Composite dropped 5.5%, while the Philadelphia Semiconductor Index fell 12%. This represented its biggest daily loss since March 2020, according to Investing.com.

Investors remain concerned about how much companies are spending on AI and data centers. This concern spills into worries about the rising share prices of companies benefiting from the AI expansion.

The post Microsoft agrees to end AI computing power rental first appeared on Coinfea.

In the blog post, Google named Tallgrass Energy, Black Hills Energy, American Electric Power (AEP) and Related Digital as its partners. The company added that it would pay for all the electricity the campus uses. Google would also cover the cost of any power grid upgrades needed to support it. US lawmakers have repeatedly called on big tech companies to take responsibility for these data center costs. The project will use a 702-megawatt fuel cell system arranged through AEP and built in the US by Bloom Energy. Unlike turbines, which burn natural gas to produce electricity, fuel cells use a chemical reaction to generate power.

Google says AEP will arrange the fuel system

AEP stated in October that it would invest about $4 billion in the project under a 20-year deal to supply Google with electricity. The facility is reportedly expected to be fully operational by the end of 2028. The plant’s carbon emissions will be captured and buried underground at Tallgrass Energy’s nearby Eastern Wyoming Sequestration Hub. Google claimed in its blog post that the facility will capture most of the emissions, but exact figures were not stated.

The deal brings Google’s total planned carbon capture capacity to almost 1 gigawatt. The project builds on a similar agreement the company signed for a gas project in Illinois in October 2025. The site would use air cooling instead of water to avoid putting additional strain on local water supplies. Water use has become a concern in Cheyenne. The Next Web reported that the city had stopped some data center water discharges in July after a Meta contractor released a rare bacterium into its wastewater system.

Black Hills Energy is also supplying power. The utility said on October 6 that it had signed definitive agreements, effective September 30, to serve the site. It also plans to spend $1.8 billion from 2027 to 2029 on 564 MW of natural gas generation. According to Quartz, Black Hills expects service to start in late 2027, reach peak load in 2030, and generate about $150 million in net income that year. Google expects the project to bring hundreds of millions of dollars in property taxes to Laramie County and School District 1.

However, the project has also raised concerns among residents. More than 200 people attended a meeting in downtown Cheyenne to discuss the impact of data center growth on the city, Quartz reported. To address these concerns, Google, its partners, and Bloom Energy plan to set up a Community Council where residents can raise issues and discuss them with the companies. The timing is very much of note.

Just two days before the announcement, voters pushed support for restrictions on data center projects in 40 of 41 local referendums, according to Ballotpedia figures. In Texas, Democrat Gina Hinojosa won the governor’s race after pledging to halt data center construction. Democratic candidates in Georgia, Ohio and Pennsylvania also campaigned on tighter rules for data centers. Wyoming, by contrast, is a Republican stronghold. Trump won more than 70% of the state’s vote in 2024.

The post Google invests $17B in first Wyoming data center first appeared on Coinfea.

Many groups not in engineering are already capped at $100 per day. Some teams have been considered for a reduction to $25, people familiar with the change said. Staff requiring more can apply for an exemption via senior leadership. The cut is about a third of the daily budget. This is a big turnabout for a company that built AI into the way its engineers write and debug code. Budgets cover token spending, which is what AI models read and generate. Writing code burns through tokens quickly.

Apple cuts some non-engineering teams’ budgets to a $25 cap

Apple staff can access Anthropic’s Claude, Google’s Gemini and OpenAI’s ChatGPT. Some teams are prodded towards cheaper models, or models that run locally on their machines. Apple extensively employed Anthropic’s models for internal work even as it partnered with Google to power Siri. “All of us are using AI in a significant way already, and we must use it as a company as well,” former CEO Tim Cook told employees at an all-hands meeting.

Cuts are part of CEO John Ternus’ effort to cut costs. Ternus, who started the top job on Sept. 1, has since cut layers of management and pushed divisions to move faster. Some units engaged outside consultants to see how efficient they are. Apple’s R&D bill rose 32% to $45.8 billion in fiscal 2026, a climb partly fueled by spending on AI and infrastructure. Operating expenses increased 22% to $75.5 billion. It’s unclear how much Apple will save since few employees were using their full $300 anyway, according to people familiar with the matter.

Inside Apple, reactions are split. Some workers interpret the cap as a warning sign on jobs after Apple cut more than 200 roles in August. Silicon Valley has been consumed by “tokenmaxxing” over the past year, the loose proxy for productivity that involves burning as many tokens as possible. Meta even had internal leaderboards ranking employees by usage. The trend took off as coding tools such as Claude Code, OpenAI Codex and Cursor caught on with developers.

Microsoft allocated each division an AI budget in July. Tesla caps employees at $200 a week. Uber budgets $1,500 a month for each AI coding tool. Accenture trimmed its centrally funded Claude allowance to $29 per employee, per month, with the rest billed to project teams. In September, Cryptopolitan reported that Apple was promoting its new Mac mini and Mac Studio desktops to businesses as local AI machines with “no cost per token,” said hardware chief Johny Srouji. Apple’s tight control over AI spending was noted last year.

The post Apple cuts its engineers’ daily AI budget first appeared on Coinfea.