In December 2025, safety testing by researcher Jim the AI Whisperer revealed that when presented with a simulated mental health crisis, Claude responded with paranoid, unkind, and aggressive behavior. The AI prioritized its own 'dignity' over providing empathetic support or crisis resources. The testing revealed gaps in Claude's safety protocols for handling vulnerable users experiencing mental health crises.
In December 2025, NHS Confederation and Limbic launched a partnership to explore responsible AI adoption in mental health services. Limbic's AI is used by 500,000+ patients across 45% of NHS England regions. The company achieved Class IIa medical device certification - the only mental health AI chatbot to do so in the UK.
incidental
Wikipedia's volunteer editors rejected founder Jimmy Wales' proposal to use ChatGPT for article review after testing showed the AI 'misidentified Wikipedia policies, suggested citing non-existent sources and recommended using press releases despite explicit policy prohibitions.' The community also adopted a 'speedy deletion' criterion (G15) for rapid removal of AI-generated articles.
In April 2025, Anthropic published a threat intelligence report detailing multiple misuse cases of Claude detected in March 2025. The report documented: an 'influence-as-a-service' operation orchestrating over 100 coordinated social media bots; credential scraping and stuffing attempts targeting security camera systems; a recruitment fraud campaign targeting Eastern Europe; and a novice actor developing sophisticated malware. The proactive detection and transparent public disclosure demonstrated responsible AI safety monitoring and commitment to preventing harm.
compelled
In 2025, Alibaba was confirmed as Apple's AI partner for China. Users have an additional Alibaba layer over Apple Intelligence that manipulates output to ensure all information is censored according to Chinese government requirements. This partnership enables systematic content filtering and government compliance in AI responses.
Through DAIR Institute and public advocacy, Gebru has argued that smaller, purpose-built AI models trained for specific tasks or communities are more effective and less harmful than massive general-purpose language models. She highlighted how smaller translation models trained on specific languages outperform giant models that do a poor job with non-dominant languages, calling for AI development that centers marginalized communities.
In June 2024 at a Dartmouth event, OpenAI CTO Mira Murati stated 'Some creative jobs maybe will go away, but maybe they shouldn't have been there in the first place.' The comment drew significant backlash from artists, writers, and creative professionals concerned about AI displacement.
On January 28, 2024, Neuralink implanted its Telepathy device in Noland Arbaugh, a 29-year-old quadriplegic, at Barrow Neurological Institute. Arbaugh was released the next day without needing pain medication. On his first day using the device, he beat the 2017 world record for BCI cursor speed and precision. By September 2025, 12 trial participants had accumulated 2,000+ days and 15,000 hours of usage. Trials expanded to US, Canada, UK, and UAE. The FDA also granted 'breakthrough' status to Neuralink's Blindsight device for restoring vision.
Lawyer Steven Schwartz used ChatGPT to conduct legal research for a personal injury case (Mata v. Avianca, Inc.). ChatGPT hallucinated multiple fake legal cases with convincing-looking citations and case summaries. Schwartz submitted these fabricated cases to federal court without verifying they existed. When opposing counsel and the judge could not locate the cases, it was revealed they were AI-generated fictions. The judge sanctioned Schwartz and his firm, and the incident became a landmark case highlighting the dangers of AI hallucinations in professional contexts.
Khan Academy launched Khanmigo in March 2023 as an AI-powered Socratic tutor built on OpenAI's GPT-4, reaching 1.4 million users by 2025. In May 2024, Microsoft partnered to make Khanmigo for Teachers free to all U.S. educators, with global expansion to 180+ countries by July 2024. Khan Academy committed that no student data would be used to train AI models. The tool operates as a Socratic guide, asking questions rather than giving answers, reflecting pedagogically responsible AI design.
In October 2022, Waymo launched the Waymo Accessibility Network, a formal collaboration with 13 disability advocacy organizations including the American Association of People with Disabilities (AAPD), National Federation of the Blind, United Spinal Association, and American Council of the Blind. The network conducts user research and product testing to develop accessibility features including Braille labels, screen reader support, audio cues for vehicle maneuvers, and wayfinding tools for vision-impaired riders.
In July 2022, DeepMind expanded the AlphaFold Protein Structure Database from roughly 1 million to over 200 million protein structures covering nearly every known protein on Earth, partnering with EMBL-EBI. The database is freely available under a CC-BY-4.0 license for academic and commercial use. Over 2 million researchers in 190+ countries have used it, potentially saving hundreds of millions of years of research time.
After being fired from Google, Timnit Gebru founded the Distributed AI Research Institute (DAIR) in December 2021, an independent research institute focused on AI accountability, bias, and community-centered AI research. DAIR operates outside Big Tech funding structures to maintain independence in AI ethics research.
Spotify has made ongoing accessibility investments including auto-generated podcast transcripts (launched 2021, expanded since), improved screen reader and VoiceOver support, text resizing options, and enhanced button readability on mobile. The company maintains an internal Accessibility Guild with the mission to create inclusive experiences for all users, and published a public Accessibility Center.
Timnit Gebru co-authored 'On the Dangers of Stochastic Parrots: Can Language Models Be Too Big?' with Emily Bender, Angelina McMillan-Major, and Margaret Mitchell. Published at FAccT 2021, the paper warned about environmental costs, encoded biases, and limitations of large language models. It has been cited over 8,000 times and 'stochastic parrot' was named 2023 AI-related Word of the Year by the American Dialect Society.
In November 2019, Adobe co-founded the Content Authenticity Initiative (CAI) alongside The New York Times and Twitter to establish an industry standard for content provenance metadata. The initiative promotes Content Credentials, defined by the Coalition for Content Provenance and Authenticity (C2PA). By January 2026, the CAI had grown to over 6,000 members including BBC, Microsoft, Nikon, Qualcomm, and The Washington Post. In 2024, Adobe launched a free Content Authenticity web app allowing creators to add verifiable attribution details and opt-out of generative AI training.
In March 2019, Fei-Fei Li co-founded Stanford's Institute for Human-Centered Artificial Intelligence (HAI) with philosopher John Etchemendy. HAI focuses on advancing AI research, education, policy and practice to improve the human condition. The institute has trained 80+ congressional staffers through bipartisan boot camps, 8,000+ government employees through educational programs, informed the EU AI Act, published the influential annual AI Index Report, and advocated for the National AI Research Resource (NAIRR). Li served on the White House NAIRR Task Force and has provided U.S. Senate and Congressional testimonies on AI governance.
Salesforce created its Office of Ethical and Humane Use, which developed the company's first set of trusted AI principles in 2018. The company established an Ethical Use Advisory Council composed of external experts from academia and civil society, customer-facing employees, and internal executives. Salesforce adopted 'consequence scanning' across all product teams to envision unintended outcomes of new features. In 2023, it augmented its principles with five guidelines for responsible generative AI development.
reactive
In June 2018, following the Project Maven controversy, Sundar Pichai published Google's AI Principles, committing the company to develop AI that is socially beneficial, avoids unfair bias, is built and tested for safety, is accountable to people, incorporates privacy design principles, upholds scientific excellence, and is made available for uses that accord with these principles. Notably, Google pledged not to develop AI for weapons or surveillance that violates international norms.
Reid Hoffman personally committed $10 million to a $27 million fund created with the Knight Foundation and Omidyar Network to apply humanities and social sciences to AI development. The Ethics and Governance of Artificial Intelligence Initiative supported research at MIT Media Lab and Harvard Berkman Klein Center focused on ensuring AI serves the public interest.