Tech giants and AI startups are expanding data collection to employee work and user content to train advanced AI agents. Elon Musk announced plans to train future Grok models on all SpaceX employee activity, while Twitch faces backlash for enabling Amazon AI training on creator streams by default. Concurrently, startups like Skan AI and Mechanize are securing major backing to monitor desktop workflows and build reinforcement learning environments, signaling an industry-wide shift toward capturing real-world human labor patterns.
Employee workplace data collection
- ▪SpaceX is exempt from the National Labor Relations Act following a February 9, 2026 National Labor Relations Board dismissal that placed the company under the Railway Labor Act, weakening employees' legal recourse to collectively challenge workplace data collection.
- ▪SpaceX CEO Elon Musk announced that the company plans to train future versions of its Grok AI model on the workplace activity, thoughts, and ideas of its approximately 14,000 to 15,000 employees.
- ▪SpaceX did not disclose details regarding which employee data will be collected, how it will be gathered, or whether workers have the option to decline participation in the Grok AI training initiative.
- ▪Meta launched its Model Capability Initiative in April 2026 to log keystrokes and screen content on US employees' work computers, but paused the program in June 2026 after an internal data leak exposed private employee conversations company-wide.
Grok 4.6 benchmark performance
- ▪Independent testing by Artificial Analysis placed Grok 4.6 at a score of 61 on its composite Intelligence Index, matching OpenAI's GPT-5.6 Sol and coming within one point of Anthropic's Claude Fable 5.
- ▪Grok 4.6 achieved a GDPval-AA v2 Elo of 1753 on agentic performance, completing knowledge-work tasks in an average of 53 turns and 0.5 billion input tokens.
- ▪SpaceXAI launched Grok 4.6 on August 12, 2026, retaining the 1.5-trillion-parameter foundation of Grok 4.5 but utilizing regenerated and filtered supervised fine-tuning trajectories to achieve performance gains.
Reinforcement learning training environments
- ▪Google is in negotiations to invest more than $1.5 billion in Mechanize, a startup founded by Tamay Besiroglu that builds virtual work environments to train and evaluate AI agents.
- ▪Scale AI reported that nearly half of its new AI training projects involve reinforcement learning environments that simulate coding, computer use, and enterprise workflows.
- ▪Uber has deployed specialized AI engineers, termed Agentic Pods, inside its corporate departments to observe employee workflows and redesign them around AI systems.
User content opt-out settings
- ▪Twitch faced user backlash after revealing that a setting allowing parent company Amazon to use creator and audience content for generative AI training is enabled by default.
- ▪Major AI platforms including OpenAI's ChatGPT, Google Gemini, Anthropic's Claude, and xAI's Grok offer user settings to opt out of having conversations used for model training, though default configurations and controls vary.
- ▪Twitch Chief Product Officer Mike Minton stated that the platform made the AI training feature opt-out because nobody would actively choose to opt in.
Desktop activity monitoring software
- ▪Skan AI's desktop software captures local screenshots, processes them to transmit anonymized metadata, and has processed over 25 billion work signals across clients including seven of the ten largest US banks.
- ▪Skan AI raised $63 million in a Series C funding round co-led by Cathay Innovation and Dell Technologies Capital to expand its enterprise desktop observation software.
Story comments
Loading comments…