AI in Medicine, Advertising, and Security: A Week of Breakthrough Decisions

A Billion Dollars for Drug Discovery

The biggest announcement of the past week remains the joint project between NVIDIA and Eli Lilly^[3]. The companies announced the creation of a human-AI collaborative laboratory for drug discovery^[1] in the San Francisco Bay area, with a budget reaching up to 1 billion dollars over five years. The NVIDIA Bio. Ne. Mo platform aims to model billions of molecular combinations before their synthesis, while Eli Lilly will contribute biological and chemical expertise. The technological partners are Thermo Fisher Scientific and Multiply Labs, responsible respectively for laboratory equipment and robotics. NVIDIA’s CEO Jensen Huang and Eli Lilly’s CEO David Ricks spoke of the chance to shorten the drug development cycle from 10–15 years to around 5–7 years, while acknowledging the risk that algorithms may suggest compounds difficult to synthesize or potentially toxic, which will require costly clinical validations.

Ads in ChatGPT and a New Revenue Model

In parallel, OpenAI announced the introduction of ads in the free version of ChatGPT^[41] alongside a new subscription plan “Go” for 8 dollars monthly^[5], which will also include ads. Users of the Plus plan (20 dollars) and Pro plan (200 dollars) will remain ad-free. The company’s communications reveal that OpenAI faces infrastructure obligations amounting to 1.4 trillion dollars over eight years, with estimated annual cash expenditures around 14 billion dollars. The company targets 20 billion dollars in recurring annual revenue by the end of 2025 and up to 125 billion dollars by 2030, assuming about 20% of that revenue will come from the generative AI advertising market. Ads in ChatGPT will appear beneath responses, clearly marked as sponsored content, while ensuring advertisers receive no direct access to conversation data.

Meta Acquires Manus and Focuses on Agents

An important move was also made by Meta Platforms, which decided to acquire the startup Manus^[29], specializing in autonomous AI agents for tasks such as flight booking, code generation, or market research. According to consistent media reports including Wall Street Journal and Reuters, the transaction is valued at over 2 billion dollars^[27], making it Meta’s third largest acquisition after Whats. App for about 19 billion dollars and Scale AI for about 14.3 billion dollars. Manus already has a market-tested product, so Mark Zuckerberg gains ready technology that can be swiftly integrated with Whats. App, Instagram, and Facebook, fulfilling the vision of “personal AI” in every app. This direction could deepen user dependency on the Meta ecosystem and accelerate further acquisitions in the agent segment.

Chinese DeepSeek Cuts Model Prices

At the model infrastructure level, attention is drawn to the Chinese Deep. Seek R1 system. This model, launched in January 2025^[31], continues to exert price pressure thanks to significantly lower computation rates. The entry price is around 0.55 dollars per million input tokens^[32], with output tokens costing 2.19 dollars, whereas a comparable OpenAI model ‘o1’ costs 15 dollars and 60 dollars respectively per million tokens. This represents up to 96% lower inference costs. According to benchmark analyses including AIME 2024 for math tasks and Codeforces for programming, Deep. Seek R1 achieves competitive results against leading US models at less than 10% of their cost. This may enforce further price cuts in the market and intensify political disputes over exporting computing power and chip design tools to China.

Agent Security and AI in Healthcare

In the area of AI agent security, the startup WitnessAI raised 58 million dollars in a Series B round^[35] led by Sound Ventures^[36], with participation from Qualcomm Ventures, Samsung Ventures, and Fin Capital. According to company data, annual recurring revenue grew by 500% year-over-year, reflecting growing demand from large enterprises for AI agent oversight tools. WitnessAI offers detection of so-called Shadow AI, user-intent-based policies, and detailed logs of agent activities. Industry analysts assess that the agent security sector could become for cybersecurity what specialized cloud protection firms were in the mid-2010s. Meanwhile, AI in healthcare is also progressing rapidly: Anthropic announced the Claude for Healthcare package^[40], including HIPAA-compliant infrastructure, integrations with electronic medical records via partner Health. Ex, beta connections with Apple Health and Android Health Connect, as well as encryption of medical data during storage and transmission.

Law, Deepfakes, and New Privacy Regulations

At the same time, the importance of regulations and ethical risks is growing. In the United States, among others, the California SB 243 law on companion chatbots^[48] came into effect, requiring clear user notification about the tool’s nature and the implementation of safety procedures in suicide risk situations, as well as California AB 2013, mandating disclosure of general training data information for models, albeit without clear enforcement mechanisms. New comprehensive data protection laws have also been adopted by Indiana, Kentucky, and Rhode Island, while Vietnam enacted its first nationwide personal data law PDPL. Abuse risks are also mounting: in Hong Kong, about 25 million dollars were fraudulently obtained using a deepfake recording of a purported CFO, and earlier FBI warnings about North Korea’s operations highlight the use of deepfakes and identity theft for infiltrating American companies. In the US, a presidential AI executive order by Donald Trump dated December 11, 2025, is already in effect, creating a federal dispute team and announcing Federal Trade Commission guidelines for March 11, 2026, while limiting states’ ability to enact separate regulations except for children’s safety and public procurement.