AI News Roundup – Anthropic barrels towards IPO with S-1 filing, Dartmouth provost resigns over AI use allegations, Chinese startup fights mosquitos using AI-enabled lasers, and more

To help you stay on top of the latest news, our AI practice group has compiled a roundup of the developments we are following.

• Anthropic has released its S-1 filing, putting the company on track for its initial public offering (IPO) later this month, according to the Financial Times. The nearly 250,000-word document, filed with the U.S. Securities and Exchange Commission, described the vast effect AI could have on the global economy, as well as a litany of “catastrophic or existential risks to humanity,” including “provid[ing] meaningful assistance to malicious actors seeking to develop chemical, biological, or other dangerous weapon[s].” The company reported rapid revenue growth in the past year, from $4.6 billion in 2025 to $33 billion through the end of September of this year. However, that growth is incredibly concentrated: just five customers accounted for a quarter of Anthropic’s 2026 revenue, and only 250 customers spent over $10 million on the company’s services. The company appears to be nearing profitability, though Anthropic has a staggering $800 billion in data center spending commitments through as late as 2045. The company is expected to list at a valuation of $2 trillion, which would place it among the most valuable companies in the world. Such a valuation would bring a windfall for Anthropic’s investors, including Amazon, which could own about 20% of the company pending conversion of convertible notes, as well as Google, which could own a 7% stake. As with SpaceX’s IPO earlier this year, insiders wishing to sell their stock will need to wait based on a staggered timeline beginning in the first few months after the IPO. Anthropic’s S-1 also proposed a novel corporate governance structure that gives retail shareholders almost no power over the company. Rather, the most powerful actor is a “Long-Term Benefit Trust,” which can elect and remove members of Anthropic’s board of directors, though Anthropic CEO Dario Amodei and his sister Daniela cannot be removed. The Amodei siblings and other cofounders of the company will retain control of the day-to-day operations of the company through their ownership of special stock that grants them 50.1% of voting power. Anthropic is expected to be offered on the Nasdaq index under the stock ticker ANT later this month.

• The Boston Globe reports on the resignation of Dartmouth College provost Santiago Schnell provoked by allegations of widespread AI usage in his personal and academic writing. The controversy began in August of this year, when Semafor reported that an op-ed written by Schnell earlier that month had been identified by the AI-detection tool Pangram (which this AI Roundup has covered in recent months) as being 100% AI-written. Schnell admitted to using ChatGPT to “refine [his] arguments” and “check for grammatical errors,” but maintained that the op-ed was based on his own thinking. The next month, reporters at Dartmouth’s student-run paper found that several opinion pieces and academic articles published by Schnell in 2026 had a median Pangram score of 96% AI-written, and Schnell’s emailed statement to the student reporters was also found to be 100% AI-written. In response to the allegations, Dartmouth convened a faculty committee to review Schnell’s work, which found that Schnell consistently used AI in his writing and was dishonest about his AI use, thus violating the college’s AI use standards. Schnell resigned before a no-confidence vote could be brought against him this month. Dartmouth President Sian Leah Beilock said that Schnell’s decision to resign was his own, but that she expects “senior leaders to hold themselves to the highest standards.” Local news outlets in New Hampshire reached out to Schnell for comment, though he referred such questions to his attorneys.

• A Chinese startup has developed an AI-powered device to fight mosquitoes using lasers, according to The New York Times. Built by the Changzhou-based company Photon Matrix, the Laser Mosquito Air Defense System tracks insects in flight and disables them by burning off their wings with high-powered lasers. The brick-sized device uses light detection and ranging (lidar) sensors, which are commonly used in aerial imaging and in self-driving cars, to emit 100,000 light pulses per second to identify airborne targets within its field of view. It then fires a Class 4 industrial laser to burn the wings off each insect, leaving it disabled. Since such lasers can cause retinal damage to the human eye in less than a second of exposure, the device incorporates multiple safeguards, including an AI camera and radar system to detect the presence of humans, pets, or reflective surfaces and prevent the beam from firing unsafely. Photon Matrix has already secured 5,000 prepaid orders for its device, which is priced at $1,000, and is preparing for mass production pending regulatory clearance in China, Europe, and the United States for sale to consumers.

• Bloomberg reports on a new poll from Pew Research that finds Americans to be open to AI usage in healthcare, though with significant safeguards. The survey of 3,488 U.S. adults conducted in June of this year revealed that while most Americans accept AI use by healthcare entities for administrative functions, they strongly resist autonomous decision-making by AI in patient care. Approximately 78% of respondents supported AI language translation for patients who do not speak English, and 76% supported AI usage in scheduling appointments, but most respondents wanted a human conducting final reviews for both tasks. 47% of respondents supported AI-assisted diagnoses, but only 3% endorsed AI making diagnoses entirely on its own. In contrast, 46% of respondents opposed any AI involvement in health insurance claim decisions. The public remains evenly divided on the general effect of AI on healthcare, though pessimism has grown sharply among young adults, where 45% now anticipate worse outcomes due to AI as compared to 29% in 2022. Significant concerns also persist regarding patient privacy and interpersonal care, with 51% expecting the security of personal health data to decline and 47% expecting patient-doctor relationships to deteriorate. The survey shows that the American public expresses far greater trust in doctors than in insurance firms, technology companies, or the federal government to safeguard patient interests and information, reflecting a broader skepticism of the nation’s institutions that has grown in the wake of the COVID-19 pandemic.

• An employee of a U.S. defense contractor allowed their personal AI coding agent to access a secure Defense Department system, according to WIRED. A worker at the consulting firm Accenture connected a secure production environment on the military’s Non-classified Internet Protocol Router Network (NIPRNet) to a personal account to use Anthropic’s Claude Code AI agent, a tool currently barred from Defense Department use amid ongoing litigation over the company’s designation by the Pentagon as a supply chain risk. To automate coding and perform other tasks, the contractor provided administrative login credentials directly to the agent, enabling it to bypass standard security controls. In doing so, the agent transmitted the contractor’s login credentials, passwords, and a network map to a third-party website previously flagged as high-risk by the National Institute of Standards and Technology due to a vulnerability on the site. NIPRNet stores controlled unclassified information, which covers sensitive subjects such as oil pipelines, nuclear submarine propulsion, and dangerous chemicals, explosives, or biological terrorism risks. The AI agent operated on the Pentagon network for two months before it was discovered and the vulnerability fixed, though sources reported that Accenture did not initially report the incident within the mandatory 72-hour window after discovery as required by federal regulations. Cybersecurity experts warned that unvetted AI-generated code tested on unclassified networks like NIPRNet could eventually be deployed inadvertently to air-gapped classified systems like the Secret Internet Protocol Router Network (SIPRNet), creating severe national security risks, especially if the code contains a vulnerability. An Accenture spokesman told WIRED that the allegations were “found to be without merit,” though did not respond to any follow-up questions or deny that an incident had occurred. The Defense Department declined to comment to WIRED on network security matters.