The AI world this week was marked by bold corporate maneuvers, fresh debates on how we measure progress, and pressing reminders of AI’s risks. From Apple’s quiet interest in snapping up leading AI startups to Anthropic uncovering hackers weaponizing generative tools, the headlines reveal both the promise and peril of this fast-moving industry. OpenAI, meanwhile, expands its footprint in India’s education sector while partnering with Anthropic on safety evaluations. Add to that the surge of graph databases riding the AI wave, and it’s clear the ecosystem is diversifying rapidly.
Graph databases surge amid AI boom: Once a niche technology, graph databases are experiencing explosive growth due to AI’s demand for complex data relationships. Unlike relational databases, graph systems model networks of entities — such as people, events, or proteins — making them ideal for powering recommendation engines, fraud detection, supply chain optimization, and knowledge graphs behind large language models. Companies from startups to enterprises are investing heavily, as graph databases become foundational for the AI-driven economy. Read more here
Apple considered buying Mistral AI and Perplexity: According to reports, Apple explored the possibility of acquiring French startup Mistral AI and U.S.-based search engine Perplexity as part of its strategy to strengthen its AI portfolio. While no deals materialized, the consideration itself signals Apple’s intent to expand beyond incremental AI features on iOS and move toward deeper control of foundation models and AI-driven search. The move would have positioned Apple directly against Microsoft’s OpenAI partnership and Google’s Gemini ecosystem. Read more here
OpenAI names ex-Coursera exec to lead India education efforts: OpenAI has appointed Raghav Gupta, former head of Coursera India, to lead its education initiatives in the region. The company aims to integrate ChatGPT into classrooms, vocational training, and language learning, targeting India’s vast and diverse student population. With Gupta’s background in scaling edtech, OpenAI hopes to navigate the unique challenges of affordability, regional languages, and teacher adoption — marking a major step in AI’s role in global education. Read more here
MIT report questions how AI progress is measured: A new MIT Technology Review report has ignited debate by suggesting that claims of AI “failure” are rooted in flawed benchmarks and expectations. The authors argue that traditional metrics, such as accuracy or benchmark leaderboards, may not reflect real-world usefulness or emergent properties of AI systems. Instead, they propose a rethinking of evaluation frameworks that capture adaptability, reasoning under uncertainty, and societal impact — elements often missed in current testing. Read more here
Google DeepMind unveils “Nano-Banana” model in Gemini: The mysterious top-rated “Nano-Banana” model that appeared in benchmark leaderboards has now been revealed as a Google DeepMind project. It has been integrated into Gemini apps and APIs under the branding “Gemini 2.5 Flash Image.” The model is designed for consistent character and object preservation in edits, solving one of the most persistent flaws in AI image generation. DeepMind claims it outperforms competitors on major image benchmarks and is optimized for speed and resource efficiency. Read more here
OpenAI and Anthropic collaborate on safety evaluations: OpenAI published findings from a joint pilot alignment evaluation exercise with Anthropic. The project tested methods for red-teaming models, measuring failure modes, and benchmarking safety frameworks. Both labs noted progress in identifying vulnerabilities, such as models’ susceptibility to harmful instructions, but acknowledged that evaluations remain incomplete and must evolve with rapidly advancing capabilities. This collaboration is part of an emerging industry-wide effort to standardize AI safety practices. Read more here
xAI open-sources Grok 2.5: Elon Musk announced that xAI has officially released Grok 2.5 as an open-source model. According to Musk, this move is part of a strategy to ensure transparency and allow developers worldwide to experiment and improve upon the model. Grok 2.5 is said to outperform its predecessor with better reasoning and multimodal capabilities, while Grok 3 is promised for release within six months. The announcement highlights the trend of major labs selectively open-sourcing to gain community support while retaining competitive advantage. Read more here
Press enter or click to view image in full size
Image credits: ChatGPT
3. Mental Health & AI Safety
AI chatbots inconsistent with suicide-related queries: A peer-reviewed study published by RAND, with support from the U.S. National Institute of Mental Health, found that leading chatbots like ChatGPT, Gemini, and Claude give unreliable and uneven responses when asked about suicide. While they often handle high-risk prompts with crisis hotlines or urgent resources, they fail more frequently on subtler, moderate-risk cues. The study stresses that such inconsistency could leave vulnerable users without timely help, raising questions about whether AI systems are ready for sensitive mental health scenarios. Read more here
Anthropic links major cyberattacks to AI-powered tools: Anthropic disclosed that a hacker group had used generative AI to conduct widespread cyber intrusions, accelerating tasks like phishing, code generation, and evasion of detection systems. While details remain limited, the company said the incident shows how malicious actors are already weaponizing AI at scale. This revelation has amplified calls for stronger AI security protocols, auditing requirements, and international coordination to prevent models from being exploited in cyberwarfare. Read more here
Press enter or click to view image in full size
Image credits: ChatGPT
4. AI Controversies & Ethical Implications
Pentagon document reveals AI propaganda ambitions: A leaked Pentagon document obtained by The Intercept reveals that the U.S. military is exploring the use of AI for propaganda, particularly to “suppress dissenting arguments” in foreign information environments. The report indicates that advanced natural language models could be used to monitor social media and insert counter-messaging at scale. Civil liberties groups have raised alarms, warning that this could normalize AI-driven manipulation and blur ethical lines in both international relations and domestic policy. Read more here
ScotRail to replace controversial AI voice on trains: ScotRail is removing its AI announcement voice known as “Iona” after backlash from passengers and the original voice actor, who alleged her voice had been cloned without fair consent. The voice, meant to modernize service, was described by travelers as “creepy” and “unnatural.” Facing growing public criticism, ScotRail confirmed it will restore human-recorded announcements, marking a rare rollback of AI deployment in public infrastructure. Read more here
Spotify in another AI music controversy: Spotify has become entangled in a fresh AI scandal after a wave of synthetic tracks sparked concerns about copyright, royalties, and disclosure. This follows earlier problems like the Velvet Sundown case, where AI-generated music was uploaded without transparency. Critics argue Spotify is not doing enough to regulate or label AI tracks, putting artists and users at risk of being misled. The platform’s handling of AI music continues to attract scrutiny from both the industry and regulators. Read more here
Coinbase CEO fires AI-averse engineers: Coinbase’s CEO Brian Armstrong reportedly terminated several engineers just a week after they joined, due to their resistance toward integrating AI into workflows. Armstrong has been outspoken about AI as a productivity multiplier, insisting that employees embrace it fully. The incident underscores growing corporate tensions where leaders demand rapid adoption of AI tools, while some workers push back against over-reliance on unproven technology. Read more here
Wikipedia study highlights “tells” of chatbot writing: Wikipedia editors and researchers have documented subtle linguistic patterns that betray AI-generated text. These include overly formal tone, unusual sentence structures, generic phrasing, and lack of genuine detail. While many chatbots can avoid obvious errors, they still leave detectable traces that make them distinguishable from human authors. The findings are important for platforms like Wikipedia that seek to filter AI content and preserve authenticity in knowledge creation. Read more here
Netflix issues generative AI guidelines for filmmakers: Netflix has published a set of official rules for how moviemakers may use generative AI in content production. The guidelines emphasize that AI can assist in pre-visualization, effects, or translation but should never replace human creativity or violate talent consent. The rules explicitly forbid using generative AI to synthesize actors’ likenesses without contracts, and they mandate clear disclosure of AI-generated elements on screen. This is Netflix’s attempt to balance technological adoption with protecting the film industry’s creative workforce. Read more here
ChatGPT predicts next 20 Super Bowl winners: In a lighthearted experiment, USA Today asked ChatGPT to forecast NFL Super Bowl champions through 2044. The AI projected repeated wins for dominant franchises like the Chiefs, Eagles, and 49ers, while also giving occasional victories to teams like the Bears, Lions, and Jaguars. Although clearly speculative, the exercise shows how generative AI is increasingly used for entertainment and fan engagement, blending predictive modeling with cultural phenomena like sports. Read more here
Outro
These stories show how AI is no longer confined to research labs or consumer gadgets — it is shaping geopolitics, education, cybersecurity, and even the underlying databases powering the digital economy. As corporations, governments, and researchers push forward, the twin themes of opportunity and oversightremain at the center of the AI debate. GT Protocol will continue tracking these shifts to help our community navigate the complex future of intelligent systems.
