Techin' Geek

Fintech firm Marquis told customers that it plans to seek compensation from its firewall provider after blaming the company for a breach that allowed hackers to steal its customers’ personal and financial data.

In a memo shared with customers this week and seen by TechCrunch, Marquis said it believes that its August 2025 ransomware attack happened because the company’s firewall service provider SonicWall had its own data breach that exposed critical security information about its customers’ firewalls. That earlier breach of SonicWall allowed hackers to obtain credentials needed to launch a ransomware attack against Marquis, the memo said.

Marquis said its third-party investigation determined that the hackers obtained information about its firewall during the breach at SonicWall, which Marquis claims was used to circumvent its firewall. Marquis confirmed in the communication that it stored a backup of its firewall configuration file in SonicWall’s cloud. 

The company was “evaluating its options” regarding its firewall provider, including the “recoupment of any expenses spent by Marquis and its customers in responding to the data incident,” according to the memo.

When reached for comment, Hanna Grimm, an agency spokesperson representing Marquis, did not address or dispute the company’s recent communication to customers, but reiterated the claim linking its breach with an earlier theft of its firewall configuration.

“In September 2025, after the data security incident affected our systems, our firewall service provider, an industry-leading cybersecurity company, publicly disclosed that a threat actor had earlier in the year gained unauthorized access to its cloud backup service,” the statement said. 

“Marquis had recently begun using this provider’s firewalls to help protect our network,” the statement added. “While the provider initially reported that fewer than 5% of customers were affected, it later clarified in October 2025 that firewall configuration data and credentials associated with all customers using the cloud backup service, including Marquis, had been accessed.”

When contacted by TechCrunch, SonicWall spokesperson Bret Fitzgerald said that the company has asked Marquis for evidence to substantiate its claims and said it would continue to engage with its customer.

“We have no new evidence to establish a connection between the SonicWall security incident reported in September 2025 and ongoing global ransomware attacks on firewalls and other edge devices,” Fitzgerald said.

The Texas-based Marquis, which allows hundreds of banks and credit unions to visualize their customers’ data, began notifying hundreds of thousands of people last month that their information was taken during its ransomware attack.

The company has access to large amounts of data belonging to consumer banking customers across the U.S., including personal information, financial data, and Social Security numbers, which the hackers stole.

SonicWall conceded in October that an earlier breach of its systems had in fact affected all of its customers who backed up their firewall files to SonicWall’s cloud. It had previously said hackers stole only a fraction of its customers’ firewall configuration files containing policies and settings.

In the communication seen by TechCrunch, Marquis said it called in a third-party to investigate whether a patch it had failed to roll out at the time of the breach could have been to blame, but concluded that the patch related to a flaw that was not exploitable in a way that could have allowed hackers to access the company’s data.

Marquis’ spokesperson declined to provide a number of how many individuals are affected by its data breach. The number of individuals known to be affected by the breach is expected to rise as new data breach notifications are submitted to state attorneys general. 

Do you know more about the Marquis data breach? Do you work at Marquis or a company affected by the breach? We would love to hear from you. To securely contact this reporter, you can reach out using Signal via the username: zackwhittaker.1337

After years of negotiations and false starts, Waymo is now allowed to operate a robotaxi service to and from the San Francisco International Airport (SFO). The Alphabet-owned company said in a blog post Thursday it will begin offering access to SFO to a select number of riders before offering it to all customers in the coming months.

Pickups and drop-offs will occur at the SFO Rental Car Center, which is accessible via AirTrain. Waymo said it plans to serve additional airport locations in the future.

Waymo’s SFO win comes as the company faces criticism and concerns about safety in some of the cities it operates. Waymo revealed Thursday that one of its robotaxis struck a child near an elementary school in Santa Monica. The National Highway Traffic Safety Administration (NHTSA) is investigating the January 23 incident, in which the child sustained minor injuries. Waymo is also being investigated by the NHTSA and the National Transportation Safety Board over the illegal behavior of its robotaxis around school buses.

Access to airports, and particularly SFO, is critical to Waymo’s business model, which hinges on geographic scale and a high volume of riders.

“Serving rides to and from San Francisco International Airport delivers one of the most requested features for our riders and further deepens our relationship with the city,” Waymo co-CEO Tekedra Mawakana said in a statement.

The company has accelerated its plans over the past year, launching into new cities, increasing its fleet size, and adding freeways to where it operates. Waymo robotaxis now service most of the San Francisco Bay Area and down into Silicon Valley, where it has access to the San Jose Airport. It also operates in parts of Atlanta, Austin, Los Angeles, Miami, and most of Phoenix, including curbside service to the Phoenix Sky Harbor International Airport.

Waymo’s push to operate at SFO has taken years. It tried and failed to secure a permit in 2023 to map SFO, a first step to bringing its robotaxis there. Waymo then rebooted negotiations with the city and airport authority and was granted a permit in March 2025 that would allow it to map SFO with some data-sharing strings attached, according to language in the agreement viewed by TechCrunch at the time.

By September, SFO and Waymo signed a testing and operations pilot permit, pushing the company closer to commercial operations at the airport.

Last year, Deezer introduced an AI detection tool that automatically tags fully AI-generated music for listeners and removes it from algorithmic and editorial recommendations.

The company announced on Thursday that it’s now making the tool available to other streaming platforms in an effort to help address the rise of AI and fraudulent streams, as well as promote transparency within the music industry and make sure human artists still get the recognition they deserve.

Alongside the move, Deezer reported that 85% of streams from fully AI-generated tracks are deemed fraudulent. Notably, the service now receives 60,000 AI tracks per day, totaling 13.4 million AI-detected songs. By contrast, in June of last year, fully AI-generated music made up 18% of daily uploads, surpassing 20,000 tracks.

​Deezer claims its AI music detection tool can identify every AI-generated track from major generative models like Suno and Udio. In addition to excluding AI-generated tracks from recommendations, Deezer’s tool demonetizes them and excludes them from the royalty pool, as the company aims to fairly compensate musicians and songwriters.

The tool’s accuracy is 99.8%, a company spokesperson told TechCrunch.

​Deezer CEO Alexis Lanternier says there has been “great interest” in the tool, and several companies have “already performed successful tests.” One such company is Sacem, the French management company that represents over 300,000 music creators and publishers, including David Guetta and DJ Snake.

​The company didn’t provide pricing information or disclose which additional companies are interested in adopting the tool. A spokesperson told us that the cost varies based on the type of deal.

​There is increasing concern about AI companies using copyrighted material to train their models, as well as about methods being used to manipulate streaming systems and commit fraud. 

One instance of music streaming fraud occurred in 2024, when a North Carolina musician was charged by the Department of Justice (DOJ) with creating AI-generated songs and using bots to stream them billions of times, resulting in more than $10 million in stolen streaming royalties. Additionally, AI bands like The Velvet Sundown have gained millions of streams.

​Bandcamp recently got fed up and banned AI-generated music altogether, while Spotify has updated its policy to address the rise of AI tracks, clarifying when AI is used in music production, reducing spam, and explicitly stating that unauthorized voice clones are prohibited on the platform.

By contrast, major record labels have resolved lawsuits with Suno and Udio, appearing to embrace AI-generated music. Last fall, Universal Music Group and Warner Music Group struck deals with these AI startups to license their music catalogs, ensuring artists and songwriters are compensated when their work is used to train AI models.

​In recent years, Deezer has taken significant steps to address concerns about AI-generated music. In 2024, it became the first music streaming platform to sign the global statement on AI training, joining actors Kate McKinnon, Kevin Bacon, Kit Harington, Rosie O’Donnell, and other notable creatives. 

​Hopefully, Deezer’s latest decision to sell its detection tool will set a precedent for other music streaming platforms to take similar actions to defend human artists and fight fraud.

paceX is reportedly lining up four major Wall Street banks for a 2026 IPO that could provide the reset the market needs. 

The company just completed a tender offer at an $800 billion valuation, and secondary market demand is through the roof. If SpaceX goes public anywhere near its rumored $1.5 trillion valuation, it could trigger an IPO cascade for other late-stage unicorns like OpenAI, Stripe, and Databricks. 

Watch as Equity host Rebecca Bellan chats with Greg Martin, Managing Director at Rainmaker Securities, about why this IPO feels different, how tech employees are cashing out through secondary markets before companies go public, and what investors are actually looking for in pre-IPO shares. 

Subscribe to Equity on YouTube, Apple Podcasts, Overcast, Spotify and all the casts. You also can follow Equity on X and Threads, at @EquityPod. 

Many in the industry think the winners of the AI model market have already been decided: Big Tech will own it (Google, Meta, Microsoft, a bit of Amazon) along with their model makers of choice, largely OpenAI and Anthropic. 

But tiny 30-person startup Arcee AI disagrees. The company just released a truly and permanently open (Apache license) general-purpose, foundation model called Trinity, and Arcee claims that at 400B parameters, it is among the largest open-source foundation models ever trained and released by a U.S. company.

Arcee says Trinity compares to Meta’s Llama 4 Maverick 400B, and Z.ai GLM-4.5, a high-performing open-source model from China’s Tsinghua University, according to benchmark tests conducted using base models (very little post training).

Like other state-of-the-art (SOTA) models, Trinity is geared for coding and multi-step processes like agents. Still, despite its size, it’s not a true SOTA competitor yet because it currently supports only text.

More modes are in the works — a vision model is currently in development, and a speech-to-text version is on the roadmap, CTO Lucas Atkins told TechCrunch (pictured above, on the left). In comparison, Meta’s Llama 4 Maverick is already multi-modal, supporting text and images.

But before adding more AI modes to its roster, Arcee says, it wanted a base LLM that would impress its main target customers: developers and academics. The team particularly wants to woo U.S. companies of all sizes away from choosing open models from China. 

“Ultimately, the winners of this game, and the only way to really win over the usage, is to have the best open-weight model,” Atkins said. “To win the hearts and minds of developers, you have to give them the best.”

The benchmarks show that the Trinity base model, currently in preview while more post-training takes place, is largely holding its own and, in some cases, slightly besting Llama on tests of coding and math, common sense, knowledge and reasoning.

The progress Arcee has made so far to become a competitive AI Lab is impressive. The large Trinity model follows two previous small models released in in December: the 26B-parameter Trinity Mini, a fully post-trained reasoning model for tasks ranging from web apps to agents, and the 6B-parameter Trinity Nano, an experimental model designed to push the boundaries of models that are tiny yet chatty.  

The kicker is, Arcee trained them all in six months for $20 million total, using 2,048 Nvidia Blackwell B300 GPUs. This out of the roughly $50 million the company has raised so far, said founder and CEO Mark McQuade (pictured above, on the right). 

That kind of cash was “a lot for us,” said Atkins, who led the model building effort. Still, he acknowledged that it pales in comparison to how much bigger labs are spending right now.

The six-month timeline “was very calculated,” said Atkins, whose career before LLMs involved building voice agents for cars. “We are a younger startup that’s extremely hungry. We have a tremendous amount of talent and bright young researchers who, when given the opportunity to spend this amount of money and train a model of this size, we trusted that they’d rise to the occasion. And they certainly did, with many sleepless nights, many long hours.” 

McQuade, previously an early employee at open-source model marketplace HuggingFace, says Arcee didn’t start out wanting to become a new U.S. AI Lab: The company was originally doing model customization for large enterprise clients like SK Telecom. 

“We were only doing post-training. So we would take the great work of others: We would take a Llama model, we would take a Mistral model, we would take a Qwen model that was open source, and we would post-train it to make it better” for a company’s intended use, he said, including doing the reinforcement learning. 

But as their client list grew, Atkins said, the need for their own model was becoming a necessity, and McQuade was worried about relying on other companies. At the same time, many of the best open models were coming from China, which U.S. enterprises were leery of, or were barred from using. 

It was a nerve-wracking decision. “I think there’s less than 20 companies in the world that have ever pre-trained and released their own model” at the size and level that Arcee was gunning for, McQuade said. 

The company started small at first, trying its hand at a tiny, 4.5B model created in partnership with training company DatologyAI. The project’s success then encouraged bigger endeavors. 

But if the U.S. already has Llama, why does it need another open weight model? Atkins says by choosing the open source Apache license, the startup is committed to always keeping its models open. This comes after Meta CEO Mark Zuckerberg last year indicated his company might not always make all of its most advanced models open source. 

“Llama can be looked at as not truly open source as it uses a Meta-controlled license with commercial and usage caveats,” he says. This has caused some open source organizations to claim that Llama isn’t open source compliant at all.

“Arcee exists because the U.S. needs a permanently open, Apache-licensed, frontier-grade alternative that can actually compete at today’s frontier,” McQuade said.

All Trinity models, large and small, can be downloaded for free. The largest version will be released in three flavors. Trinity Large Preview is a lightly post-trained instruct model, meaning it’s been trained to follow human instructions, not just predict the next word, which gears it for general chat usage. Trinity Large Base is the base model without post-training.

Then we have TrueBase, a model with any instruct data or post training so enterprises or researchers that want to customize it won’t have to unroll any data, rules or assumptions.

Acree AI will eventually offer a hosted version of its general release model for, it says, competitive API pricing. That release is up to six weeks away as the startup continues to improve the model’s reasoning training.

API pricing for Trinity-Mini is $0.045 / $0.15, and there is a rate-limited free tier available, too. Meanwhile, the company still sells post-training and customization options.