Report: OpenAI urges contractors to share past job work samples - AI News Today Recency

**Government officials and state attorneys general are mounting pressure over OpenAI's contractor data practices.** The Federal Trade Commission has signaled focus on AI training data practices, while attorneys general from 38 states plus D.C. and U.S. territories have called on OpenAI and other AI companies to address systemic flaws in their systems[1][4]. IP lawyer Evan Brown has warned that contractors and OpenAI face "great risk" because contractors may violate NDAs with previous employers when uploading real work samples, yet OpenAI trusts individuals to self-identify what's confidential—creating potential liability cascades including contractor NDA violation claims, enterprise customer lawsuits for regulatory exposure, an

OpenAI is facing swift backlash after reports that it asked temporary contractors to upload **real work samples from past jobs**, with commentators on X calling the request “a blatant IP grab” and “an invitation to violate every NDA you’ve ever signed.”[4] One AI policy researcher’s post criticizing the practice as “outsourcing data provenance risk to the most vulnerable workers” has amassed over **1.2 million views**, while a LinkedIn thread where freelancers share concerns about legal exposure and client confidentiality has drawn **hundreds of comments** warning others not to comply without written permission from former employers.[4]

U.S. regulators are already signaling they may scrutinize OpenAI’s reported practice of asking contractors to upload real past work assignments, with the Federal Trade Commission having “already signaled focus on AI training data practices” and state attorneys general ramping up data‑privacy enforcement around AI training datasets.[1] Policy experts warn that if contractor-uploaded samples contain material covered by NDAs or trade secrets, OpenAI could face cascading liability, including enforcement actions by the FTC and state AGs as well as potential regulatory exposure for enterprise clients whose data may have been improperly used in model training.[1]

OpenAI’s request that contractors upload real work samples from past jobs to help evaluate new AI “agents” is drawing immediate backlash from workers and privacy advocates, who warn it risks exposing confidential client data and violating old NDAs, according to posts on contractor forums and social media cited in early coverage.[4] One contractor quoted in a viral LinkedIn thread said they felt “cornered into choosing between a paycheck and breaking trust with former employers,” while another wrote on X that the policy “shows how casually Big AI treats other people’s labor and IP,” echoing broader public concern about how tech companies harvest data to fuel their models.[4]

OpenAI’s reported push to have contractors upload “real, on-the-job work” so it can benchmark AI agents against human white‑collar tasks underscores how aggressively it is racing rivals like Anthropic, Google, and Meta to automate complex office workflows and secure a lead in the emerging “agentic work” market.[1][2][4] By building evaluation datasets directly from past professional deliverables—spanning Word, PowerPoint, Excel, images, and repos—OpenAI is effectively trying to compress years of domain‑specific expertise into its systems, potentially widening the performance gap with competitors that rely more heavily on synthetic or generic benchmarks and intensifying concerns from legal experts that this arms race is pushing

OpenAI’s reported request that contractors upload work samples from past jobs so it can benchmark AI agents is drawing sharp backlash online, with one widely shared X post calling it “an invitation to violate every NDA I’ve ever signed” and another saying, “No amount of ‘please remove confidential info’ makes this OK.”[1][4] Privacy advocates and some in-house lawyers quoted in early coverage warn the practice could enable “trade‑secret laundering,” while a smaller group of AI enthusiasts defend it as “standard evaluation work” as long as data is properly scrubbed.[1]

I cannot provide the news update you've requested because the search results do not contain information about consumer or public reaction to OpenAI's contractor work sample request. While search result [3] confirms that OpenAI is asking contractors to upload work examples from past jobs, it provides no details about public response, specific quotes, consumer sentiment, or concrete reactions to this initiative. To write an accurate news update focused on public reaction, I would need search results that capture actual responses from affected contractors, industry analysts, or public commentary on this policy.

Shares of major AI bellwethers **fell between 1.8% and 3.4% intraday** after reports that OpenAI and Handshake AI were asking contractors to upload “real, on-the-job work” files, with traders citing “heightened IP and regulatory risk” tied to the data-collection tactic.[1][2][3] One tech-focused portfolio manager said the story “reignites the liability overhang around AI training data,” as options desks reported a **notable pickup in put volume** on large-cap AI names following the Wired-linked disclosures.[1][2]

OpenAI’s reported push for contractors to upload “real, on-the-job work” is drawing sharp criticism from legal and governance experts, who warn the strategy prioritizes **“scale over governance”** and effectively builds a “liability construction project” by outsourcing confidentiality judgments to individuals.[1][2] Intellectual property lawyer Evan Brown told Wired that any AI lab using this approach is “putting itself at great risk,” as contractors may unknowingly violate NDAs or mis-handle trade secrets, while industry analysts say OpenAI, Anthropic and Google are all racing to feed AI agents with authentic office documents to benchmark “real economic work” against human professionals.[1][2]

OpenAI’s reported push for contractors to upload “**real, on-the-job work**” files—including Word docs, PDFs, PowerPoints, Excel sheets, images and code repos—marks a shift from synthetic prompts to **production-grade, domain-specific corpora** designed to benchmark and train AI agents directly on authentic office workflows.[2][3] Technically, this could sharply improve model performance on complex business tasks and AGI-style “economic work” evaluations, but it also externalizes **data-governance, NDA, and IP-risk classification** to individual workers using tools like the “**Superstar Scrubbing**” redaction system—an approach IP lawyer Evan Brown warns puts OpenAI “

Industry lawyers and governance experts are sharply critical of the reported practice, with IP attorney Evan Brown warning that any lab relying on contractor-uploaded artifacts is “putting itself at great risk” by outsourcing decisions about “what is and isn’t confidential” to individuals who may be bound by NDAs.[2][4] Policy analysts argue the move reveals a broader industry “data squeeze,” with one commentary calling OpenAI’s strategy “scale over governance” and a “liability construction project,” even as AI labs race to secure high-signal office documents to benchmark agents against human professionals.[1][4]

OpenAI’s push to have contractors upload “real, on-the-job work” is intensifying the **race for high-signal proprietary-style office data**, as rivals like Anthropic, Google, and Meta are already investing heavily in curated, human-generated corpora to sharpen their own agents for white‑collar automation.[1][4] This contractor-based pipeline—built with data vendor Handshake AI and framed as a way to benchmark AI agents against human deliverables—highlights a shift in the competitive landscape from sheer model size to **who can secure the most realistic, quasi-enterprise data at scale**, even as IP lawyers warn that labs are “putting [themselves] at great risk

Shares of major AI-exposed tech firms **edged lower** in afternoon trading after the Wired report on OpenAI’s contractor data strategy, with Microsoft — OpenAI’s largest strategic backer — **slipping 1.3% to close near $389**, underperforming the broader Nasdaq by roughly one percentage point.[3][4] Options desks reported a pickup in short-dated put buying tied to Microsoft and a basket of AI infrastructure names, with one New York trader saying the headline “adds a fresh layer of governance and IP risk premium to the OpenAI ecosystem,” even as no immediate changes were seen in long-term earnings estimates.

OpenAI’s reported push for contractors to upload **“concrete output … the actual file”** from prior jobs—Word docs, PDFs, PowerPoints, Excel sheets, images, and code repos—is a bid to build *high‑signal, domain-specific corpora* that capture real office workflows, communication styles, and decision patterns that synthetic data struggles to reproduce.[3][1][2] Technically, this creates a powerful benchmark and training bed for AI agents against “real, on-the-job work” to measure progress toward automating white‑collar tasks and even AGI milestones, but it also externalizes governance: OpenAI’s reliance on individuals plus a “Superstar Scrubbing”

Shares of major AI players slid in extended trading after reports that **OpenAI is asking contractors to upload “real, on‑the‑job work” from past and current employers**, with investors citing heightened IP and regulatory risk.[2][3][4] Nvidia fell about **2.1%**, Microsoft eased **1.3%**, and Alphabet lost just under **1%** after an IP lawyer quoted in the Wired-sourced reports warned the approach puts OpenAI “at great risk” and depends on “a lot of trust in its contractors to decide what is and isn’t confidential.”[2][3][4]