Did OpenAI Fire Its Whistleblowers?

OpenAI announced it has parted ways with three researchers, Jasmine Wong, Tongue Tomec, Corbach, and Makita Balsani, due to violations of its policies on handling sensitive information. The company reports that these individuals mishandled confidential data outside established procedures, which broke trust within the organization. OpenAI confirmed the termination and stated the investigation verified misuse of sensitive info outside protocol. Sources say the misconduct involved sharing information with an AI safety organization, which has sparked controversy. One of the fired employees, Corbach, was involved in Redwood research and had served as OpenAI’s technical contact for investigations related to the Hugging Face incident, where OpenAI models were hacked. The firing has led to online speculation about whistleblowing, with some suggesting the leaked incidents may have been justified disclosures. Experts and commentators emphasized the need for greater transparency from AI organizations, especially regarding safety testing and handling internal investigations.

Cloudflare introduced two multimodal decision models, Clef and Clef Flash, which they claim are smarter and faster than the recently popular Jeb model. Built on Quinn 3.827B and Quinn 3.59B, respectively, these models can answer structured questions like yes/no, multiple choice, and rankings, while also supporting images and video, and a 64K token context window. Cloudflare reported that Clef outperforms Jeb in accuracy and speed, according to their internal benchmarks, though these scores are not yet independently verified. The models are available via Cloudflare’s Workers AI platform, making them fast through edge computing, and they are open-weight under Apache 2.0 licenses on Hugging Face. Despite some reported slower speeds compared to other open models, Clef’s strength lies in its multimodal capabilities and competitive accuracy, aiming to challenge the dominance of models like Jeb in decision-making tasks.

Suno has announced a new feature called Speech, now in public beta, which generates human-like spoken voices from scripts or prompts, with optional background music. This expansion marks Suno’s move beyond its AI music generation roots, aiming to diversify the platform amid ongoing lawsuits related to AI-generated music. The tool allows users to choose script descriptions or precise text for voiceovers, with customization options for gender, style, and voice variety. The feature supports up to eight-minute audio tracks and is designed for diverse uses from poetry to dramatic voiceovers. Suno’s CEO highlighted the potential of combining speech with music, although they acknowledge the current limitations, such as occasional pronunciation issues. This new addition offers a flexible structure for creating engaging audio content, even as the quality continues to improve through user feedback.

The Dutch eyewear retailer Hans Anders has suspended sales of Meta’s Ray-Ban smart glasses in the Netherlands and Belgium due to increasing privacy concerns. Similar restrictions are being considered in the UK, where Wetherspoons has limited the glasses’ use in its pubs. The controversy focuses on the glasses' built-in cameras, which have an always-on indicator light and cannot be fully turned off, raising fears over covert recording. These concerns come amid ongoing lawsuits and regulatory warnings about the privacy implications of smart glasses. Meta and Luxottica, the maker, defend the product, stating that the glasses' cameras disable if tampered with. Privacy groups and online watchdogs argue that such devices threaten individuals’ privacy, pushing policymakers to scrutinize these technologies more closely.

Following President Trump’s speech at the UN where he proposed renaming AI as 'super intelligence' or SI, there was a sudden spike in domain registrations with the .si suffix in Slovenia. Official data shows that in September, 44,000 new .si domains were registered compared to only about 2,000 in August, a rise of over 2,100%. The spike is attributed, at least partly, to curiosity or agreement with the rebranding, as claimed by Slovenia’s .si domain registry. Trump’s initiative to rebrand AI in this way was part of a broader move to emphasize the importance and legitimacy of the technology, although it also sparked satire and skepticism, especially after the incorrect spelling of his signature on official documents. The increased domain activity signals a mix of political branding and public interest in superintelligence.

Bloomberg reports that a California man was arrested for allegedly smuggling over $300 million worth of restricted Nvidia AI chips to China via Malaysia and Singapore, despite US export controls. This highlights ongoing vulnerabilities in export enforcement and the extensive black market for AI hardware, with estimates of hundreds of thousands of chips crossing borders illegally. Probes suggest that China may have obtained enough AI hardware to fill data centers, raising concerns about national security and export regulations. Meanwhile, Vanity Fair features an in-depth profile of Larry Ellison, Oracle’s billionaire founder. Sketched as a self-taught programmer and relational database pioneer, Ellison’s career is also characterized by his competitive bravado, public feuds with Bill Gates, and obsession with yacht racing. His personality and business trajectory exemplify the archetype of a tech magnate blending innovation with flamboyant personal pursuits.