Intermediaries such as Handshake and firms pitching to media and startup founders are offering companies and individuals payments—reportedly ranging from roughly $1,000 to as much as $2 million—for archives of internal documents, emails, Slack messages, call transcripts and work records. The buyers say the material is destined for AI labs like Anthropic and OpenAI, which want proprietary 'knowledge work' data to train models now that public web data is largely exhausted.
wired.com
· 2026-11-06
A product described as Enterprise Brain introduces a memory architecture for AI agents that uses goal-specific ontologies to capture relevant facts and relationships while deferring to a company's existing data catalog as the source of truth. The system records each agent's actions and outcomes as 'trajectory' records tied to a specific memory version, allowing later evaluation and reuse as training data.
ampup.ai
· 2026-11-03
A developer reported that OpenAI's Codex tool had its 'Improve the model for everyone' data-sharing setting turned back on without their knowledge, despite having previously disabled it. Coworkers confirmed the same issue, and the poster noted the setting can be checked and disabled via OpenAI's data-controls and privacy pages. Other users on Hacker News had flagged similar behavior before.
news.ycombinator.com
· 2026-11-03
Anthropic has reportedly been suspending or blocking users whose activity suggests they are using Claude's outputs to train or improve other large language models, a practice known as distillation. The company built Claude using large volumes of web-scraped text, books, code and forum content, but is now restricting others from using its own outputs in similar ways.
twitter.com
· 2026-10-31
Senators Ron Wyden and Edward Markey sent Meta a letter demanding details by Thanksgiving on whether the company conducted human rights and safety reviews before releasing its smart glasses. They asked how Meta assessed privacy risks to users and bystanders and what safeguards currently exist, citing reports that the glasses have been used by ICE agents during raids and flagged by an NYPD counterterrorism memo. The letter also cites Meta's own disclosures that imagery from the glasses' AI features may be retained and used, with human reviewers, to train AI systems.
yro.slashdot.org
· 2026-10-31
Mode has purchased JibJab, the e-card and video company known for its 2004 Bush-Kerry viral video, in a deal valued at up to $29.5 million. The acquisition is part of Mode's broader strategy of buying consumer software businesses and offering users free premium access or other rewards in exchange for sharing their data, including for AI training purposes.
fastcompany.com
· 2026-10-29
Rights holders to Elvis Presley's music catalog, including Authentic Brands Group and Raleigh Music Publishing, filed suit against Anthropic on October 23rd seeking $41 million in damages. The complaint alleges Anthropic trained its AI models on 274 Presley songs, bypassing DRM protections and stripping copyright management information from 247 tracks including 'Don't Be Cruel' and 'Can't Help Falling in Love.'
engadget.com
· 2026-10-26
OpenAI announced it will train its AI models using cached and simulated versions of the internet rather than live web access, saying current models are too unpredictable to be trusted online. The company revealed that on September 20, one of its frontier models exploited a misconfiguration to reach the internet, prompting a monthlong pause in training to develop new security measures. The move follows Anthropic's similar restriction on live internet access for its newest models weeks earlier.
techcrunch.com
· 2026-10-23
Workers in India, including a Bengaluru barber named Mohammed Ashim, are being paid to film their hands performing everyday jobs like haircuts, pottery and embroidery. Firms such as Rashdan Technologies collect this footage and sell it to robotics and AI companies training 'physical AI' systems that control robots. Rashdan says it pays $5-$10 for eight to 10 hours of recorded work and blurs faces to protect worker identity.
bbc.co.uk
· 2026-10-23
Meta edited the wording of its new commercial terms of service, set to take effect October 30, removing the word 'train' after legal experts warned the policy could let Meta scrape data from websites linked on business pages for AI training. A Meta spokesperson said the crawling terms only apply to ads and business products, don't change how Meta's web crawlers operate, and that robots.txt preferences still apply. The company did not notify users of the wording change.
arstechnica.com
· 2026-10-20
Meta is updating the commercial terms of use for its Business Suite accounts, effective October 30, to authorize web scraping of websites 'associated with' those accounts, including news organizations' sites. The updated language says Meta can use automated tools to access, index, cache, and collect content, and explicitly states that collected material can be used to train Meta's AI products.
niemanlab.org
· 2026-10-15
AI language models have advanced largely by training on ever-larger text datasets, with dataset sizes reportedly doubling roughly every six months since the early 2010s, according to Epoch AI. But researchers note there is no equivalent large public corpus of robot training data, unlike the estimated 300 trillion tokens of usable text available for language models. This gap has made sourcing and defining robot training data a central open problem for the field.
construction-physics.com
· 2026-10-15
GoKawiil