Private conversations with xAI’s chatbot Grok have been exposed online, raising serious concerns over user privacy and AI safety. Forbes found that Grok’s ‘share’ button created public URLs, later indexed by Google and other search engines.
The leaked content is troubling, ranging from questions on hacking crypto wallets to instructions on drug production and even violent plots. Although xAI bans harmful use, some users still received dangerous responses, which are now publicly accessible online.
The exposure occurred because search engines automatically indexed the shareable links, a flaw echoing previous issues with other AI platforms, including OpenAI’s ChatGPT. Designed for convenience, the feature exposed sensitive chats, damaging trust in xAI’s privacy promises.
The incident pressures AI developers to integrate stronger privacy safeguards, such as blocking the indexing of shared content and enforcing privacy-by-design principles. Users may hesitate to use chatbots without fixes, fearing their data could reappear online.
Would you like to learn more about AI, tech and digital diplomacy? If so, ask our Diplo chatbot!
Communication, empathy, and judgment were dismissed for years as ‘soft skills‘, sidelined while technical expertise dominated training and promotion. A new perspective argues that these human competencies are fundamental to resilience and transformation.
Researchers and practitioners emphasise that AI can expedite decision-making but cannot replace human judgment, trust, or narrative. Failures in leadership often stem from a lack of human capacity rather than technical gaps.
Redefining skills like decision-making, adaptability, and emotional intelligence as measurable behaviours helps organisations train and evaluate leaders effectively. Embedding these human disciplines ensures transformation holds under pressure and uncertainty.
Career and cultures are strengthened when leaders are assessed on their ability to build trust, resolve conflicts, and influence through storytelling. Without funding the human core alongside technical skills, strategies collapse, and talent disengages.
Would you like to learn more about AI, tech, and digital diplomacy? If so, ask our Diplo chatbot!
Chief of Microsoft AI, Mustafa Suleyman, has urged AI firms to stop suggesting their models are conscious, warning of growing risks from unhealthy human attachments to AI systems.
In a blog post, he described the phenomenon as Seemingly Conscious AI, where models mimic human responses convincingly enough to give users the illusion of feeling and thought. He cautioned that this could fuel AI rights, welfare, or citizenship advocacy.
Suleyman stressed that such beliefs could emerge even among people without prior mental health issues. He called on the industry to develop guardrails that prevent or counter perceptions of AI consciousness.
AI companions, a fast-growing product category, were highlighted as requiring urgent safeguards. Microsoft AI chief’s comments follow recent controversies, including OpenAI’s decision to temporarily deprecate GPT-4o, which drew protests from users emotionally attached to the model.
Would you like to learn more about AI, tech and digital diplomacy? If so, ask our Diplo chatbot!
Researchers at the University of California, Davis, have revealed that generative AI browser assistants may be harvesting sensitive data from users without their knowledge or consent.
The study, led by the UC Davis Data Privacy Lab, tested popular browser extensions powered by AI and discovered that many collect personal details ranging from search history and email contents to financial records.
The findings highlight a significant gap in transparency. While these tools often market themselves as productivity boosters or safe alternatives to traditional assistants, many lack clear disclosures about the data they extract.
Researchers sometimes observed personal information being transmitted to third-party servers without encryption.
Privacy advocates argue that the lack of accountability puts users at significant risk, particularly given the rising adoption of AI assistants for work, education and healthcare. They warn that sensitive data could be exploited for targeted advertising, profiling, or cybercrime.
The UC Davis team has called for stricter regulatory oversight, improved data governance, and mandatory safeguards to protect users from hidden surveillance.
They argue that stronger frameworks are needed to balance innovation with fundamental rights as generative AI tools continue to integrate into everyday digital infrastructure.
Would you like to learn more about AI, tech and digital diplomacy? If so, ask our Diplo chatbot!
Imagine dreaming of your next holiday and feeling a rush of excitement. That emotion is when your attention is most engaged. Neuro-contextual advertising aims to meet you at such emotional peaks.
Neuro-contextual AI goes beyond page-level relevance. It interprets emotional signals of interest and intent in real time while preserving user privacy. It asks why users interact with content at a specific moment, not just what they view.
When ads align with emotion, interest and intention, engagement rises. A car ad may shift tone accordingly, action-fuelled visuals for thrill seekers and softer, nostalgic tones for someone browsing family stories.
Emotions shape memory and decisions. Emotionally intelligent advertising fosters connection, meaning and loyalty rather than just attention.
Would you like to learn more about AI, tech and digital diplomacy? If so, ask our Diplo chatbot!
Google has patched a high-severity flaw in its Chrome browser with the release of version 139, addressing vulnerability CVE-2025-9132 in the V8 JavaScript engine.
The out-of-bounds write issue was discovered by Big Sleep AI, a tool built by Google DeepMind and Project Zero to automate vulnerability detection in real-world software.
Chrome 139 updates (Windows/macOS: 139.0.7258.138/.139, Linux: 139.0.7258.138) are now rolling out to users. Google has not confirmed whether the flaw is being actively exploited.
Users are strongly advised to install the latest update to ensure protection, as V8 powers both JavaScript and WebAssembly within Chrome.
Would you like to learn more about AI, tech, and digital diplomacy? If so, ask our Diplo chatbot!
A new study reveals that prominent AI models now show a marked preference for AI‑generated content over that created by humans.
Tests involving GPT‑3.5, GPT-4 and Llama 3.1 demonstrated a consistent bias, with models selecting AI‑authored text significantly more often than human‑written equivalents.
Researchers warn this tendency could marginalise human creativity, especially in fields like education, hiring and the arts, where original thought is crucial.
There are concerns that such bias may arise not by accident but by design flaws embedded within the development of these systems.
Policymakers and developers are urged to tackle this bias head‑on to ensure future AI complements rather than replaces human contribution.
Would you like to learn more about AI, tech and digital diplomacy? If so, ask our Diplo chatbot!
Google Translate may soon evolve into a full-featured language learning tool, introducing AI-powered lessons rivalling apps like Duolingo.
The latest Translate app release recently uncovered a hidden feature called Practice. It enables users to take part in interactive learning scenarios.
Early tests allow learners to choose languages such as Spanish and French, then engage with situational exercises from beginner to advanced levels.
The tool personalises lessons using AI, adapting difficulty and content based on a user’s goals, such as preparing for specific trips.
Users can track progress, receive daily practice reminders, and customise prompts for listening and speaking drills through a dedicated settings panel.
The feature resembles gamified learning apps and may join Google’s premium AI offerings, though pricing and launch plans remain unconfirmed.
Would you like to learn more about AI, tech and digital diplomacy? If so, ask our Diplo chatbot!
Meta is launching a research lab focused on superintelligence, led by Scale AI founder Alexandr Wang, in an attempt to regain ground in the global AI race.
Mark Zuckerberg is reportedly in talks to invest billions into Scale, reflecting strong confidence in Wang’s data-driven approach and industry influence.
While Meta’s past efforts with its Llama models gained traction, its latest release, Llama 4, failed to meet expectations and drew criticism.
Wang’s appointment arrives during an ongoing talent exodus from Meta, with several senior AI researchers departing for rivals or founding startups.
The new lab is separate from Meta’s existing FAIR division, led by Yann LeCun, who has dismissed the idea of chasing superintelligence. Meta’s partnership with Scale mirrors deals by Microsoft, Amazon, and Google, aiming to secure top AI talent without formal acquisitions.
Would you like to learn more about AI, tech and digital diplomacy? If so, ask our Diplo chatbot!
A new study from Arizona State University researchers suggests that chain-of-thought reasoning in large language models (LLMs) is closer to pattern matching than accurate logical inference. The findings challenge assumptions about human-like intelligence in these systems.
The researchers used a data distribution lens to examine where chain-of-thought fails, testing models on new tasks, different reasoning lengths, and altered prompt formats. Across all cases, performance degraded sharply outside familiar training structures.
Their framework, DataAlchemy, showed that models replicate training patterns rather than reason abstractly. Failures could be patched quickly through fine-tuning on small new datasets, but this reinforced the pattern-matching theory.
The paper warns developers against relying on chain-of-thought reasoning for high-stakes domains, emphasising the risks of fluent but flawed rationale. It urges practitioners to implement rigorous out-of-distribution testing and treat fine-tuning as a limited patch.
The researchers argue that applications can remain effective for enterprise use by systematically mapping a model’s boundaries and aligning them with predictable tasks. Targeted fine-tuning then becomes a tool for precision rather than broad generalisation.
Would you like to learn more about AI, tech, and digital diplomacy? If so, ask our Diplo chatbot!
