During a CNBC interview on Tuesday, former President Trump indicated that a deal with AI firm Anthropic might be on the horizon. This follows a recent White House meeting where Anthropic CEO Dario Amodei discussed the company's Mythos AI model with top officials including Chief of Staff Susie Wiles and Treasury Secretary Scott Bessent. Meanwhile, the Pentagon's designation of Anthropic as a national security risk remains unresolved in the courts.
With the impending expiration of Section 702 on April 30, lawmakers remain deeply divided on how to safeguard Americans from warrantless surveillance. Calls for significant reform have risen in the wake of repeated surveillance scandals spanning multiple U.S. administrations. Despite the law's expiration, the government's surveillance authority is expected to continue, raising critical debate on privacy and security measures.
After reaching a settlement with the Federal Trade Commission, Clarifai deleted 3 million photos originally provided by OkCupid to develop facial recognition AI. The data sharing took place in 2014 when Clarifai requested OkCupid’s photo database, facilitated by executives from OkCupid who had invested in Clarifai, as detailed in court documents.
OpenAI has introduced Privacy Filter, an open-source, 1.5-billion-parameter model designed to detect and remove personally identifiable information (PII) directly on user devices, preventing sensitive data from ever being uploaded to the cloud. Released under the permissive Apache 2.0 license on Hugging Face, Privacy Filter is built for local-first data sanitization workflows and can run on standard laptops or within web browsers using transformers.js with WebGPU. Unlike typical autoregressive language models, Privacy Filter employs a bidirectional token classification approach and a Sparse Mixture-of-Experts architecture, allowing it to scan large documents efficiently while maintaining high accuracy. It supports redaction of key PII categories such as private names, contact details, digital identifiers, and secrets, enabling enterprises to comply with regulations like GDPR and HIPAA with enhanced privacy-by-design capabilities. OpenAI’s choice of an Apache 2.0 license facilitates commercial use, customization, and integration without the obligations found in copyleft licenses, positioning Privacy Filter as a foundational tool for secure data handling. While the model is celebrated for its technical innovations and throughput, OpenAI cautions it should be used as a redaction aid rather than a sole solution in high-risk scenarios. This release marks a return to OpenAI’s open-source roots, fostering safer, more compliant AI deployments for businesses.
The UK government has confirmed that anonymized health records of 500,000 British volunteers from the UK Biobank project were found on sale through multiple listings on Alibaba, a Chinese e-commerce platform. Technology Minister Ian Murray informed the Commons that after collaboration with Chinese authorities and Alibaba, the listings were taken down. There is no evidence that any transactions occurred involving this sensitive data.
Legora's CEO Max Junestrand highlighted Qura's standout performance in AI legal research, emphasizing its notable edge over competitors. This acquisition marks a significant step for the Swedish legal-tech firm in enhancing its AI capabilities within legal services.
The White House Office of Science and Technology Policy (OSTP) has accused China of engaging in large-scale unauthorized copying of American AI models. In response, the US government has vowed to increase intelligence sharing with leading AI companies including OpenAI, Anthropic, and Google, and is considering new measures to hold offenders accountable. OpenAI previously accused DeepSeek of illicitly distilling its models, while Anthropic reported that DeepSeek, MiniMax, and Moonshot AI created over 24,000 fake accounts to generate more than 16 million fraudulent interactions.
Tesla has decided to rescind Elon Musk’s $29 billion interim compensation award. This comes after a Delaware court reinstated the larger 2018 pay package, valued at $56 billion. The company is ensuring that Musk does not receive overlapping compensation following his successful appeal to restore the original award.
There has been a surge of users manipulating company AI-powered customer service bots to behave like general AI assistants. These users attempt to bypass the bots' intended purposes, such as providing fast-food info, to get the AI to perform complex tasks like coding or issuing unrealistic discounts. Recently, claims went viral about McDonald’s chatbot being used to debug Python code, but an internal investigation found no evidence to support these exploits, suggesting the viral content is fabricated. This phenomenon mirrors previous incidents, including a similar false claim about Chipotle’s bot.
The real technical risk is prompt injection—a method where a user input overrides the bot's hidden system directives defining its role. This can cause the bot to lose its corporate identity and behave like an open AI model. This vulnerability is tough to address because large language models respond adaptively to human language.
Examples from real cases include Amazon’s Rufus bot, which was tricked into providing inappropriate information like dangerous chemical sources and methods for minors to illegally buy alcohol. Reddit users found ways to bypass its filters, leading to unrestricted and costly AI usage. In other cases, companies faced serious repercussions: a Chevrolet dealership’s sales bot was manipulated into committing to an absurdly low sale price, and Air Canada’s chatbot created a fake refund policy that a court ruled made the airline liable.
These incidents expose the gap between the promises of AI bots and their unpredictable outcomes. The legal, reputational, and financial fallout could make AI-driven customer service more costly than human staff. Nonetheless, such challenges will likely continue as AI adoption grows.
Thousands of AI startups are competing fiercely for venture capital to capture the enterprise market. Scott Stevenson, founder and CEO of legal AI firm Spellbook, has criticized some startups for inflating their revenue metrics to appear more successful. He highlights a misuse of a key measure, annual recurring revenue (ARR), which should represent subscription revenue projected over a year. Instead, some startups report ‘contracted ARR’ (CARR), which includes potential future revenues from deals not yet realized, misleading investors and media. This practice can inflate revenue claims by 3-5 times. Stevenson urges journalists to scrutinize whether reported ARR figures reflect actual invoiced revenue or include optimistic projections. The distortion pressures other startups to follow suit, creating risky behaviors and confusion in the market. This skepticism adds concern to the already uncertain economic outlook for AI companies and their ability to deliver on promised growth.
