AI companies pay $30K for work documents
Anthropic watermarks AI output
Anthropic admits AI suffers from a crisis of trust
Build AI Human Agents that Support, and Scale Your Business For You
Big companies like Amazon, Salesforce, and Deloitte are putting AI video agents on their websites. Now you can too, without a developer and a huge budget. A PAL is an AI video agent you can build in Tavus PAL Maker: it can see, hear, and talk with your customers face-to-face in real time. Describe what you want your PAL to do in a few sentences, say, “a friendly front-desk helper that books appointments and answers FAQs,” and Tavus will build it for you. Deploy it to your site in a few clicks. It handles questions 24/7, speaks 50+ languages, and remembers repeat visitors. You can even upload a custom face to build brand awareness. Build a PAL today and see what your storefront’s been missing. Build a human edge for your brand, your customers will thank you. Use code SBA for 50% off!
AI companies pay $30K for work documents
Technically, it doesn’t need your help. It just needs your emails, your internal messages with coworkers, and any documents you’ve written on the job. Call it the next stage in the education of AI. Having largely exhausted the open internet — scraping billions of web pages and paying humans to rate chatbot responses — AI companies are now hunting for a new kind of training data: the kind generated inside actual workplaces.
The push is playing out on multiple fronts. Elon Musk recently told SpaceX employees that their internal data would be used to train Grok, the AI chatbot built by his company xAI. “It will inherit your thoughts and ideas,” Musk told staff at a recent all-hands meeting. Google, meanwhile, paid $10 million for corporate data from bankrupt Spirit Airlines, according to court filings. It wasn’t the only bidder: Mercor, an AI training startup, offered $7.5 million for the same trove.
Now Handshake AI is taking a more direct approach. The AI training firm is offering professionals up to $30,000 to contribute “high-quality written documents” — a bid to buy the kind of workplace writing that internet scraping can’t easily capture. To qualify, applicants must have worked in fields such as consulting, finance, law, software engineering, or data science. Handshake AI said contributors must own the documents outright and be authorized to share them. It’s the latest sign of just how hungry the AI industry has become for training data — and how far companies are willing to go, and pay, to get it….
Anthropic watermarks AI output
Anthropic said it will begin embedding invisible watermarks into text generated by its Claude AI models, part of a broader push for transparency as AI-written content becomes harder to distinguish from human writing.
Starting with models launched on or after August 2, Claude will weave what the company calls an “imperceptible watermark” directly into the text it generates. The mark won’t be visible to readers and won’t change the meaning, quality, or readability of Claude’s response, Anthropic said in a post to its help center.
Because the watermark is embedded in the words themselves rather than added afterward, it will travel with the text when copied and pasted elsewhere, and may persist through some editing. Anthropic cautioned, however, that a heavy rewrite could strip it out entirely — meaning the presence of a watermark, or its absence, isn’t a definitive proof of AI involvement one way or the other.
The move is tied to Anthropic’s compliance with Article 50 of the European Union’s AI Act, whose transparency obligations took effect August 2, 2026. But the company isn’t limiting the rollout to Europe: markings will cover output from supported Claude models everywhere the tools are used — including the Claude API, Claude Code, Claude Cowork, and Claude Tag. Anthropic said it also plans to provide third parties with tools to detect them.....
Anthropic admits AI suffers from a crisis of trust
Dario Amodei, the co-founder and chief executive of Anthropic, took to X over the weekend to push back against critics who blame him for stoking public anxiety about artificial intelligence — a rare foray into social media for an executive who typically avoids it. In a lengthy post published Saturday, Amodei rejected the idea that he bears outsized responsibility for the public’s growing unease about AI, though he acknowledged the industry faces genuine trust problems.
He did concede one point critics have long raised: that AI, by its nature, tends to concentrate power. But Amodei argued this isn’t a byproduct of regulation, as some skeptics claim. Instead, he pointed to AI scaling laws — the principle that a model’s capabilities improve as the computing resources and data used to train it increase. That dynamic, he said, naturally favors organizations with access to the most computing power. Open-weight models, which allow broader access to AI systems, offer only a partial fix, he added, since they simply redistribute the advantage toward whoever controls the most chips and computing infrastructure.
Amodei also addressed criticism that his own public statements have been excessively alarmist. He disputed the characterization, even though he has previously made headlines with stark predictions — most notably his forecast that AI could eliminate up to half of all entry-level white-collar jobs. More recently, however, he has softened that framing….