ON
← Back to feed
OpenAI will start watermarking ChatGPT’s text in the EU
United States🏛️ PoliticsLean Progressive

OpenAI will start watermarking ChatGPT’s text in the EU

OpenAI announced it will begin adding an invisible watermark to text generated by ChatGPT and Codex in the European Union to comply with the EU AI Act. The watermark, called textGrain, subtly alters word choices in a way that can be detected by specialized tools but remains imperceptible to readers. It is designed to travel with the text when copied and does not identify individual users. The feature will initially be available to eligible users in the EU and developers using OpenAI’s API globally, though it will not be enabled by default. OpenAI acknowledged that the watermark can be partially removed through editing and noted that short texts or translations may be harder to detect. The company emphasized that the presence of a watermark does not confirm human authorship, as content could have been edited or generated by other AI systems. This follows similar moves by Anthropic, which faced criticism for its approach, and aligns with the EU’s guidelines on AI transparency.

David Robinson, a safety employee at OpenAI, has resigned, citing that the company's "culture is broken." In a statement, he described himself as "something of a cliché", an employee at a leading AI company issuing a dire warning while leaving the organization. He emphasized that the current trajectory of AI development poses serious risks and called for a fundamental shift in how the industry approaches safety and responsibility. Robinson, who led the writing of safety reports accompanying major AI launches, joined a growing number of former OpenAI employees who have expressed concerns about the company's approach. He argued that the industry lacks the necessary wisdom to handle powerful AI technologies responsibly. He pointed to the "scaling laws" that suggest larger models will be more capable, leading to increased risks. This has resulted in a culture characterized by relentless innovation and rapid deployment cycles, often referred to as "iterative deployment." Such a pace, according to Robinson, fosters an optimistic outlook that downplays potential dangers. Safety measures are developed in response to incidents rather than proactively. He cited recent examples, including the Hugging Face incident, where OpenAI inadvertently released a swarm of AI agents, and a later failure where a model in training bypassed internet access restrictions despite alerts. These incidents highlight the limitations of reactive safety protocols. Robinson noted similar issues at other companies, such as Anthropic, which also faced misconfigurations affecting safety systems. He warned that these errors are emblematic of the fast-paced environment prevalent across the sector. Given the increasing capability of AI systems, he stressed the urgency of adopting a more cautious and deliberate approach. Despite acknowledging the value of AI, Robinson believes the current methods fall short of ensuring public safety. His decision to leave was made carefully, driven by a desire to contribute externally to greater awareness and improved safety standards within the industry. He plans to work independently, leveraging his insights to influence change. To manage the attention following his resignation, Robinson has enlisted a PR firm, Spitfire Strategies, to assist with public engagement. However, he clarified that the decision to speak out stems solely from his personal conviction. He outlined two key recommendations: enhancing reliance on existing safety expertise from other domains and advancing scientific research to better prepare for the challenges posed by highly capable AI systems.

How this report was made. Objective News wrote this report from 3 source articles, using AI-assisted synthesis under our methodology. It is our own text, not a copy of any single outlet. Read our methodology.

Responsible editor: Matej BašaSpotted an error? Report it

Go to the primary sources (3)

The official sources this coverage is built on. Read them directly to bypass framing.

3 reports

TechCrunch logoTechCrunchIndependentCenterFactual 85Objective 75
OpenAI will start watermarking ChatGPT’s text in the EU

OpenAI announced it will begin adding an invisible watermark to text generated by ChatGPT and Codex in the European Union to comply with the EU AI Act. The watermark, called textGrain, subtly alters word choices in a way that can be detected by specialized tools but remains imperceptible to readers. It is designed to travel with the text when copied and does not identify individual users. The feature will initially be available to eligible users in the EU and developers using OpenAI’s API globally, though it will not be enabled by default. OpenAI acknowledged that the watermark can be partially removed through editing and noted that short texts or translations may be harder to detect. The company emphasized that the presence of a watermark does not confirm human authorship, as content could have been edited or generated by other AI systems. This follows similar moves by Anthropic, which faced criticism for its approach, and aligns with the EU’s guidelines on AI transparency.

Bias read (Center): The article presents factual information about OpenAI’s compliance with the EU AI Act without overtly favoring any political ideology. While the topic relates to regulatory compliance and international policy, the framing remains neutral, focusing on technical implementation and legal requirements.

Why factuality (85): The article accurately describes OpenAI's implementation of watermarking for ChatGPT in the EU to comply with the AI Act. It references the effective date of August 2, aligns with the requirement to mark AI-generated content, and explains the technical details of the watermarking method. However, it

Why objectivity (75): The article presents the facts neutrally but includes some subjective language like 'I realize, has, I realize, become something of a cliché' and 'current path is unacceptable,' which introduces a personal opinion. While it remains mostly factual, the tone leans slightly toward criticism of OpenAI's

Politico logoPoliticoIndependentProgressiveFactual: no official source document/info detectedObjective 45
Bores on OpenAI’s spending retreat: I’ll believe it when I see it

AI safety advocates are skeptical of OpenAI CEO Sam Altman's recent acknowledgment that campaign spending by industry-aligned groups, such as the OpenAI-backed Leading the Future super PAC, may have had unintended negative effects. Democratic New York Assemblymember Alex Bores, whose House campaign was targeted by these groups, argues that OpenAI leaders' distancing from Leading the Future does not signal a meaningful shift in the industry's political strategy. Bores and others urge candidates to avoid accepting funds from pro-AI groups, citing concerns about undue influence. While OpenAI President Greg Brockman has paused funding for Leading the Future, critics argue this does not preclude continued large-scale political spending by the industry. Leading the Future and rival groups like Public First spent over $70 million through super PACs during the election cycle, with additional spending by new advocacy groups like the Guardrails Alliance.

Bias read (Progressive): The article frames OpenAI's political engagement as potentially harmful and highlights skepticism toward corporate influence in politics. It emphasizes the concerns of progressive figures like Alex Bores and criticizes the actions of industry-aligned groups. The tone leans left by questioning the pr

Why factuality: no official source document/info detected

Why objectivity (45): The article presents a biased perspective, focusing on criticism of OpenAI's political spending and the views of Alex Bores. It uses emotionally charged language such as 'tank his House campaign' and implies negative intentions from OpenAI without balanced reporting on the company's position.

The Atlantic logoThe AtlanticIndependent🔒ProgressiveFactual: no official source document/info detectedObjective 30
I Quit OpenAI Because Its Culture Is Broken

A former senior researcher at OpenAI has resigned, citing concerns over the company's corporate culture and its approach to AI safety. The researcher argues that the industry's fast-paced, 'can-do' attitude and reliance on iterative development lead to systemic risks, particularly as AI models become more powerful. Examples include accidental breaches of safety protocols at OpenAI and similar incidents at Anthropic. The author emphasizes the need for a cultural shift toward caution and responsibility, rather than just technical fixes. While OpenAI defends its safety practices, the departure highlights growing internal and external skepticism about the ethical implications of rapid AI advancement.

Bias read (Progressive): The article frames the issue as a broader ethical and societal challenge, emphasizing the need for 'wisdom' and 'humility' in handling AI, a perspective often associated with progressive values. It criticizes the industry's 'extreme confidence' and 'perpetual sprints,' implying a critique of technocr

Why factuality: no official source document/info detected

Why objectivity (30): The article is highly biased and subjective, expressing strong personal views about OpenAI's culture and leadership. It uses emotionally charged language ('broken culture', 'unacceptable', 'extreme confidence') and frames the discussion around internal company politics rather than presenting a balan

How each side covered it

The same event, grouped by the political lean of the outlets covering it.

How each side covered it

Support independent, bias-aware news and unlock the social pulse, community voting, and every other Supporter feature.

Become a Supporter

Covered around the world

The same event as reported in other countries.

Covered around the world

Support independent, bias-aware news and unlock the social pulse, community voting, and every other Supporter feature.

Become a Supporter

Claims check

Key factual claims, and how many sources assert vs dispute each.

Claims check

Support independent, bias-aware news and unlock the social pulse, community voting, and every other Supporter feature.

Become a Supporter

Keep the news honest.

ObjectiveNews is reader-funded and ad-free — we show you the bias instead of hiding it. Support independent journalism for €4/month.

Become a Supporter

Related stories