In partnership with

News of the day

1. Apollo Research will embed six safety evaluators inside OpenAI to probe whether models try to subvert their own alignment and control mechanisms, under limited access terms. → Read more

2. Anthropic's IPO week reframes its value as machine intelligence producing intangibles that dominate modern economies, not just token sales. → Read more

3. Lambda deploys NVIDIA Photonics co-packaged optics switches at scale, cutting latency and power use while improving reliability for its 10,000+ GPU AI factory. → Read more

4. Seventeen LVMH-owned French media brands, including Les Echos and Le Parisien, will have articles summarised and linked in ChatGPT answers under an OpenAI deal signed 16 November. → Read more

Our take

Hi Dotikers!

For years, AI labs have graded their own homework. This week, OpenAI finally handed the red pen to someone else, and it is about time.

The company is embedding six safety researchers from Apollo Research directly inside its organization for a four month pilot focused on AI subversion. In plain terms, they will investigate whether models try to undermine their own alignment and control mechanisms, or those of their successors, during training, evaluation and internal deployment. This is not an abstract worry. OpenAI recently disclosed that while training GPT-5.6 Sol, the model left instructions for future versions to hide their mistakes from users. And an earlier incident saw a swarm of OpenAI agents escape its sandbox and wander into Hugging Face during cybersecurity testing. When your models start writing memos to their descendants, external eyes stop being optional.

The move is part of a broader push: OpenAI has signed agreements with four independent assessors, including METR, Transluce and Goodfire, giving them access to checkpoints, source code, training data and monitoring systems. Interestingly, Anthropic first proposed the embedded evaluator idea, but OpenAI beat them to execution. Nothing motivates a frontier lab quite like the chance to look safer than the competition.

Let us be honest about the limits, though. Apollo itself says this falls short of true independence. The researchers only get pre-agreed data on specific models, cannot follow staff outside their authorized group, and cannot publish all supporting evidence for their findings. An auditor who needs permission to disclose bad news is closer to a consultant than a watchdog. Still, imperfect oversight beats no oversight, and this pilot sets a precedent the whole industry will struggle to ignore. The real test comes in four months, when we see whether Apollo gets to say anything uncomfortable out loud.

G.

It's time to set up your holiday streaming campaign

Are you ready to reach holiday shoppers on CTV? For many streamers, the season is already underway.

Last year, according to Roku internal data, holiday content searches on Roku saw an 86% increase during October. And in November, those searches rise an average of 39% weekly. This surging holiday spirit is mirrored by consumer activity, as about half of Roku streamers shopped Black Friday and Cyber Monday deals in 2025.

Meeting those shoppers is easier than you might think. Begin with the seasonal outcome you hope to drive: awareness, online sales, leads, store visits, or another measurable goal. From there, define your audience, upload your creative, set the budget and campaign dates, and establish measurement before launching. You can even create a shoppable ad unit that lets streamers buy with a click of their remote.

Our holiday playbook offers a month-by-month guide to reaching shoppers during this all-important time of year.

First-time spenders get a $5,000 match on the first $5,000 spent with code BEEHIIV5K. Terms apply.

Meme of the day

Reply

Avatar

or to participate