The U.S. Justice Department charged Google software engineer Michele Spagnuolo with insider trading, alleging the employee made $1.2 million trading on Polymarket based on confidential business information.

Spagnuolo, who used the name “AlphaRaccoon” on Polymarket, has worked at Google for over 12 years, according to information on LinkedIn.

“As alleged, Spagnuolo violated the duties he owed to his employer and used Google’s confidential business information to make more than $1.2 million in trading profits on Polymarket,” Jay Clayton, the United States Attorney for the Southern District of New York, said in a press release. “Insider trading compromises the integrity of our markets, and the American people want this greed-driven conduct investigated and prosecuted.”

Prediction markets like Polymarket, Kalshi, and others allow users to bet on pretty much anything. Insider trading is not allowed on these platforms because it’s illegal, but some users still commit the offense. The Justice Department recently charged a U.S. Army soldier for allegedly using his insider knowledge of the U.S. military operation to capture Venezuelan president Nicolás Maduro to make $400,000 on Polymarket.

According to the complaint, Spagnuolo risked over $2.7 million on wagers related to Google’s 2025 Year in Search, a marketing campaign in which Google reveals the world’s most popular searches of the year. Spagnuolo allegedly accessed confidential, internal Google Search data about the most-searched celebrities to inform his bets.

“Polymarket worked closely with the U.S. Attorney’s Office for the Southern District of New York and the CFTC, and is the only prediction platform to date whose cooperation has led to insider trading charges in the United States,” a Polymarket spokesperson told TechCrunch. “Blockchain trading is transparent, traceable, and bad actors leave footprints. We are committed to maintaining accurate, fair, and transparent markets as well as enforcing our rules and working with our regulators and law enforcement.”

A Google spokesperson told TechCrunch the company is working with law enforcement on its investigation.

“The employee accessed our marketing material using a tool available to all employees, but using such confidential information to place bets is a serious breach of our policies,” Google said in an emailed statement, “We’ve placed the employee on leave and will take the appropriate action.”

How many Ps are in Google? According to Google, there are two.

There’s also is also “exactly 1 ‘r’ in the word ‘poop’,” Google’s AI Overview says, as well as two ‘d’s in the word journalism, yet spelled it: j-o-u-r-n-a-d-i-s-m. Google did at least identify that there is one P in the last name of the U.S. president, but spelled it as t-r-p-u-m.

You didn’t need to be a prophet to predict that Google’s AI-forward Search overhaul was going to go over poorly. We’ve done this before. The first time Google added AI Overviews to Search, the feature ended up citing satirical posts from The Onion and Reddit, advising people to eat rocks and put glue on their pizza.

This time around, as Google doubles down on its commitment to make generative AI the centerpiece of its 29-year-old flagship product, it’s not surprising to see it stumble.

“Counting within words has been a known challenge for LLMs, and we’re working to fix this particular issue,” Google told TechCrunch in an emailed statement.

These basic spelling errors may seem familiar. LLMs, the kind of artificial intelligence that powers chatbots and other text-generators, are not built to understand spelling. It’s been a running joke for years that whenever a company unveils a new AI model, you should ask it how many ‘r’s are in the word strawberry. These AI models — which can code an app in seconds, or solve problems that have stumped mathematicians for decades — are about as good as a kindergartener at spelling.

Google’s AI overview woes reach beyond silly spelling mistakes though. Google already patched an issue from last week in which searching the word “disregard” would yield what looked like a dictionary definition of the word, only the definition was shown as, “Understood. Let me know whenever you have a new prompt or question!” But these spelling errors have remained amusing because they’re so difficult to quash.

As researchers have previously explained when we’ve asked about these spelling conundrums, AI doesn’t perceive sentences as units of language made up of words and letters. Many LLMs are built on transformers models, which break down text into tokens, which can be full words, syllables, or letters, depending on the model. Instead of “reading” like a human would, the AI converts the text into numerical representations of itself, which are then contextualized to help the AI come up with a logical response.

“LLMs are based on this transformer architecture, which notably is not actually reading text. What happens when you input a prompt is that it’s translated into an encoding,” Matthew Guzdial, an AI researcher and assistant professor at the University of Alberta, told TechCrunch. “When it sees the word ‘the,’ it has this one encoding of what ‘the’ means, but it does not know about ‘T,’ ‘H,’ ‘E.’”

The token-based architecture that powers LLMs like Google’s AI overview is inherently limiting, and researchers haven’t been optimistic that they can solve the spelling problem.

“It’s kind of hard to get around the question of what exactly a ‘word’ should be for a language model, and even if we got human experts to agree on a perfect token vocabulary, models would probably still find it useful to ‘chunk’ things even further,” Sheridan Feucht, a PhD student studying large language model interpretability at Northeastern University, told TechCrunch. “My guess would be that there’s no such thing as a perfect tokenizer due to this kind of fuzziness.”

This isn’t necessarily an urgent problem on researchers’ minds, since the utility of LLMs doesn’t come in their capacity to spell. But these blatant failures help us remember that AI is not perfect, even if it may sometimes seem like an all-knowing power beyond our comprehension. We cannot blindly trust AI outputs without double-checking their accuracy.

Triomics, a startup building an AI-powered platform to help oncologists and administrative staff automate data-heavy tasks like clinical trial matching and appointment prep, has raised $22 million in Series B funding.

The round was led by Battery Ventures, with participation from returning backers Nexus Venture Partners, Lightspeed, Y Combinator, and others.

The good news is that oncology breakthroughs are keeping patients alive longer.  That welcome trend, however, is creating dense, multi-year medical records that take healthcare staff a long time to review and decipher.

A typical medical chart includes physician progress notes, imaging and pathology reports, and even scans of faxes. “We have seen medical records [with] thousands of pages of information,” Triomics co-founder Sarim Khan (pictured left) told TechCrunch.

Founded in 2021, the startup raised a $15 million in Series A in mid-2024. Initially focused on helping doctors identify the most suitable clinical trials for their patients, Triomics expanded its platform as LLM capabilities grew. Over the last couple of years, Triomics added verifiable patient summaries to its platform, surfacing key information directly inside the tools clinicians already use, without requiring them to switch applications.

By reducing appointment prep time, these summaries give oncologists more time with their patients. The efficiency gain matters beyond individual appointments: in oncology, where patient histories are unusually complex and staff burnout is a persistent problem, tools that reduce administrative load have an outsized impact.

Triomics is also used to automate the tedious task of submitting tumor reports to government registries, a legal mandate for cancer centers.

While generic AI agents excel at basic summaries, prominent institutions like Memorial Sloan Kettering (MSK) and Yale Cancer Center use Triomics because its models are trained specifically on oncology data, Khan explained.

Triomics most direct competition comes from AI medical scribes like Abridge and Microsoft’s Nuance — tools that use AI to listen to and document patient-doctor conversations — when it comes to summarizing patient charts.

Despite the fierce competition, Triomics is growing fast. According to Khan, the startup expanded its enterprise customer base fourfold over the past year, driving a 10-fold increase in annualized recurring revenue.