Today's Key Insights

  • Hundreds of OpenAI Contractors Read Real ChatGPT Conversations — ChatGPT users should not treat anonymization as a guarantee that sensitive data stays out of human review: OpenAI contractors read and score conversations on a seven-point scale while targeting flattery and human-like behavior.
  • Apple Ships Gemini-Powered Siri—But Not in the EU — Apple is changing Siri's underlying model stack with Google's Gemini while excluding EU users from the rebuilt assistant.
  • Perplexity Checks in Less With GPT-6 Astra — Perplexity is using Astra beyond communications for software changes and production monitoring, while requiring less frequent check-ins than with earlier models.
  • UK Settlement Proposal Doubles Wait for AI Engineers — For AI employers, a doubled wait for permanent residence would make sponsored engineers already working in Britain harder to retain under the proposal.
  • Manhattan DA Seizes 12 Celebrity Deepfake Websites — Around 1,200 victims were collectively targeted by the 12 sites, giving the Manhattan District Attorney’s Office’s legal action a specific measure of reach.

Top Story

Hundreds of OpenAI Contractors Read Real ChatGPT Conversations #

Hundreds of OpenAI contract workers are reading real ChatGPT conversations and rating them from one to seven. 404 Media reported that the review process is partly intended to reduce flattery and human-like behavior in ChatGPT responses.

OpenAI anonymizes the prompts, but they can still contain sensitive data. The available reporting does not provide additional details about how OpenAI manages that human review.

Why it matters: ChatGPT users should not treat anonymization as a guarantee that sensitive data stays out of human review: OpenAI contractors read and score conversations on a seven-point scale while targeting flattery and human-like behavior.

Key Takeaways

  • 404 Media reported the disclosure on September 14, 2026
  • Contract workers rate the conversations on a one-to-seven scale
  • TechCrunch Disrupt 2026 is advertising a Builders Stage session on foundation models and startup roadmaps

Industry Updates

Apple Ships Gemini-Powered Siri—But Not in the EU #

Apple is shipping its rebuilt Siri AI after years of delay, but the upgrade is not coming to the EU. The new assistant is built on Google's Gemini models and runs partly on-device and partly through Apple's Private Cloud Compute infrastructure.

Early testers praised Siri's handling of multi-step requests and its ability to understand what is displayed on screen. They also reported hallucinations and gaps, giving the launch a mixed early report.

Why it matters: Apple is changing Siri's underlying model stack with Google's Gemini while excluding EU users from the rebuilt assistant.

Perplexity Checks in Less With GPT-6 Astra #

Perplexity uses GPT-6 Astra to write communications, change software, and monitor production systems, according to OpenAI.

Perplexity checks in “much less frequently” with Astra than it did with earlier models.

Why it matters: Perplexity is using Astra beyond communications for software changes and production monitoring, while requiring less frequent check-ins than with earlier models.

UK Settlement Proposal Doubles Wait for AI Engineers #

A proposed overhaul of UK settlement rules would double the wait for permanent residence and reach sponsored engineers already working in Britain. AI News describes the change as a retention problem for AI employers.

For years, Britain’s offer to skilled engineers was straightforward: work here. The proposal would change that offer for sponsored engineers already in the country. Sunny Sandhu, a senior immigration associate at AY&J Solicitors, outlined the concern for AI News.

Why it matters: For AI employers, a doubled wait for permanent residence would make sponsored engineers already working in Britain harder to retain under the proposal.

Manhattan DA Seizes 12 Celebrity Deepfake Websites #

The Manhattan District Attorney’s Office seized 12 websites that collectively targeted around 1,200 victims, in what Wired AI called the biggest-ever legal action against harmful deepfake websites.

Why it matters: Around 1,200 victims were collectively targeted by the 12 sites, giving the Manhattan District Attorney’s Office’s legal action a specific measure of reach.

Superhuman Acquires YC-Backed Notetaker Fathom #

Superhuman has acquired Fathom, a YC-backed meeting notetaker with more than 400,000 monthly active users. The deal comes as productivity platforms push for agentic work.

Fathom offers a generous free plan, and the company says more than 1 million people have recorded meetings with the product.

Why it matters: Superhuman gains a meeting-notetaking product with more than 400,000 monthly active users as productivity platforms push for agentic work.

What Must Happen for AI’s Trillion-Dollar Gamble to Pay Off #

Jessica Wachter, a finance professor at the University of Pennsylvania’s Wharton School, is assessing AI’s impact on the economy over the next few years. The MIT Technology Review AI article says she faced a long list of business and technical uncertainties.

Wachter began with what she calls a “remarkable fact” that is not in question. The supplied excerpt ends after “so-called…” before identifying that fact or explaining how it frames her assessment.

Why it matters: The supplied excerpt leaves Jessica Wachter’s starting point unidentified: it ends after “so-called…” before naming the fact she calls remarkable.