Kevin Neary's Linkedin Analytics

Reasoning is likely the biggest gap between theory and practice when it comes to AI Agents.

🔵 AI Literacy Isn’t Optional — It’s a Strategic Imperative According to Elon University's 2025 report Being Human in 2035, 61% of global AI experts expect the shift in human capacity and behavior to be deep and meaningful — even fundamentally revolutionary. 🧠 The report warns that: - Empathy, deep thinking, agency, and purpose may decline. - Trust in shared values and moral judgment could erode. - Overreliance on AI may shrink human confidence and metacognition. - Yet amidst the disruption, curiosity, decision-making, and creativity are expected to grow — if we equip ourselves with the tools to lead. 🎯 Tangible Takeaway for Leaders: Start Future-Fit Skill Mapping Now The report emphasizes that empathy, ethics, and adaptability will be core leadership capabilities by 2035. But the call to action is clear: 🛠 Begin actively mapping current roles to future human-AI collaboration skills today. Not just technical AI skills, but also: - Cognitive adaptability - Emotional intelligence - Human-AI interface fluency 🧭 What you can do: - Run an AI-readiness audit across leadership teams - Embed AI literacy and ethical leadership into your 2025 upskilling roadmap - Create a culture of ongoing future skills experimentation 🔑 This is why AI Literacy for Leaders is no longer a soft skill — it's a governance mandate, a compliance necessity, and a competitive differentiator. 🎯 In the EU, it’s also now a legal obligation. The EU AI Act formally recognizes the need for AI Literacy across management layers — not just for technical teams, but also in boardrooms and leadership circles. As AI becomes embedded in our identities, decisions, and social structures, the question is no longer “Will this affect me?” — it's “Are we ready to lead responsibly through it?” 👉 That’s why we’re building the next generation of Responsible AI Leaders — through strategic training, practical tools, and governance frameworks. If your leadership team is still asking “why AI literacy?” — this report just gave you 286 pages of reasons. Summary below. #AILeadership #ResponsibleAI #EULaw #AICompliance #M&A #DigitalGovernance #Exits #AIEthics #AI2035 #AITransformation #AIForLeaders #Marketing

🔍 AI Agents Are Here—But Are We Ready? Agentic AI systems aren’t just the next frontier in automation. They’re intelligent systems with autonomy, memory, tool access—and increasingly, the power to act independently. That means real value for business: ✅ Workflow orchestration ✅ Developer productivity ✅ Always-on sales and HR support ✅ Strategic knowledge reuse But that autonomy comes with real risk. And most organizations aren’t prepared. From this week’s deep dive into IBM’s new whitepaper on agentic AI, here’s what stood out: 📌 AI agents can generate misleading outputs (hallucinations) during function calls. 📌 Open-ended tool use can lead to unpredictable outcomes—and even ethical breaches. 📌 Human dignity, trust, and autonomy are directly at stake when systems act without visibility. 🚨 In short: Just because an AI agent can act doesn't mean it should. ✅ Governance frameworks ✅ Transparent architecture ✅ Risk detection models ✅ Human-in-the-loop oversight …aren’t “nice to have.” They’re survival strategies. 💡 Business recommendation? Don't wait until you're deploying agents to build trust. Start now: Document your current workflows and decision rights. Identify agent use cases that touch customer data or regulatory processes. Assign AI governance focal points before your first implementation. I’m working with organizations building this foundation—if you’re heading toward agentic AI, let’s connect. #AgenticAI #ResponsibleAI #Governance #Automation #AILeadership #AIReadiness

This is an important step forward. The ability to consistently classify AI incidents is key to learning from them and ultimately preventing future harm. Great to see the MIT AI Risk Repository taxonomies now integrated into the AI Incident Database. A valuable contribution to the Responsible AI ecosystem.

Adversarial AI Is Already Here: 10 Things Every AI Leader Should Know from the Latest NIST Report. 1. This isn’t theoretical anymore. Real-world adversarial attacks are happening. NIST cites over 80,000 attempts to bypass face recognition systems for fraud, using masks, deepfakes, and wigs​ 2. Most models aren’t robust enough. NIST warns that current mitigation techniques for AI attacks are often bypassed. Many defenses fail against stronger, adaptive attacks​ 3. There’s no one-size-fits-all defense. Three main classes of mitigations stand out: adversarial training, randomized smoothing, and certified defenses, but each has trade-offs like accuracy, cost and complexity​ 4. Predictive AI (PredAI) vs. Generative AI (GenAI) attacks need different playbooks. The report distinguishes attacks by AI type. For example, GenAI is more vulnerable to prompt injections and hallucinations while PredAI is more at risk from data poisoning and evasion​ 5. Your LLMs can be tricked. GenAI-specific threats include prompt injection and indirect manipulation of context windows. These aren’t just bugs they’re risks to be managed proactively​ 6. “Clean-label” attacks are dangerous and hard to detect. Attackers can manipulate training data with labels that appear normal, making detection and prevention extremely challenging​ 7. No common language = more risk. One of NIST’s main goals is to standardize the language of adversarial machine learning (AML). Misalignment across AI and cybersecurity teams creates vulnerabilities 8. Your AI system is only as secure as its weakest interface. The report details how APIs, training pipelines and deployment endpoints are all potential entry points for adversaries​ 9. Formal methods are coming, for critical sectors. NIST suggests formal verification approaches (used in aviation and cryptography) may soon be required for AI in critical systems. But it will be expensive​ 10. You need an AML strategy. Now. NIST isn’t just categorizing attacks, they’re preparing future standards. That means regulators will follow. And companies without a mitigation framework will fall behind. Takeaway: AI security and responsible deployment go hand-in-hand. If you’re building, buying or auditing AI systems, adversarial risk is no longer optional to understand. - Start with AI risk assessments - Align AI and cybersecurity teams - Review your model lifecycle for exposure points - Stay up to date with evolving standards like NIST’s AML taxonomy Want help decoding the NIST report into practical AI governance actions? Let’s connect. #ResponsibleAI #TrustworthyAI #AIStandards #AdversarialML #NIST #AICompliance #AIsecurity #LLM #AIleadership Source: Adversarial Machine Learning A Taxonomy and Terminology of Attacks and Mitigations

In healthcare, it's primarily a question of trust.

AI transformation - I agree Jan, a balanced strategy—some hiring, some partnerships and yes lots and lots of upskilling - starting with Ai Literacy training across the organization to build a solid foundation and address the elephant in the room "alignment".

Prompt Engineering Evolves Responsibly: Key Insights from OpenAI’s GPT-4.1 Latest Prompting Guide OpenAI’s GPT-4.1 marks a leap forward in instruction-following and steerability, but with great power comes great responsibility. Here’s what AI practitioners should keep in mind to promote Responsible AI and AI Literacy: 1. Clarity and Specificity Empower Control GPT-4.1 follows instructions more literally than before, making clear, explicit prompts essential. This precision reduces ambiguity, helps prevent unintended outputs, and supports ethical use by minimizing misinterpretation or bias amplification. 2. Structured Prompts Enhance Transparency Using well-organized formats like Markdown, bullet points, or XML not only improves model understanding but also makes AI reasoning more interpretable to users. This transparency is a cornerstone of AI literacy, helping users grasp how AI arrives at answers. 3. Iterative Reasoning and Planning Foster Reliability Encouraging the model to “think step-by-step” or outline plans before acting improves output quality and reduces hallucinations. This aligns with responsible AI principles by promoting verifiable, trustworthy responses rather than guesswork. 4. Explicit Boundaries and Ethical Guardrails Prompting the model to avoid certain behaviors or to admit uncertainty (“I don’t know”) helps mitigate risks like misinformation or overconfidence. Combined with content moderation and bias testing, these guardrails uphold fairness and safety. 5. Educating Users on Prompt Engineering AI literacy involves teaching users how to craft effective prompts and understand model limitations. OpenAI’s guide emphasizes that prompt engineering is empirical and iterative, encouraging continuous learning and evaluation to responsibly harness GPT-4.1’s capabilities. By adopting these principles, AI developers and users can unlock GPT-4.1’s full potential while fostering ethical, transparent, and informed AI interactions. #ResponsibleAI #AILiteracy #GPT4 #PromptEngineering #EthicalAI

For Health AI - scaling often means localizing

McKinsey & Company’s latest explainer on AI agents shows how these tools are moving beyond chatbots and copilots into multi-agent ecosystems capable of learning, planning and acting. From a #Responsible_AI and #AI_Literacy standpoint - here are five key takeaways: 1. AI agents need governance, not just guidance. As agents begin orchestrating decisions and coordinating with other agents, ethical oversight becomes essential. These systems can escalate decisions, correct outputs or self-improve, but without clear rules and accountability, their actions may create unseen risks. 2. Bias and trust must be designed in, not audited later. The report calls out mechanisms like “critic agents” and “bias checkers” that work in tandem with core agents. These should be part of system design from the start - not bolted on later. Trust is earned through transparency and reliability. 3. AI agents will change organizational culture. Like the shift from manual to digital decades ago, adopting AI agents is not just a tech upgrade - it’s a structural one. Leadership must prepare for cultural transformation, new team structures, and training to support AI-first operating models. 4. Regulatory readiness is not optional. With the EU AI Act and other frameworks emerging, organizations will need policies on agent transparency, data use, escalation paths and role boundaries. AI agents interacting with customers or staff must comply with evolving standards. 5. Leadership literacy will make or break AI deployments. The most successful organizations will not just deploy agents, they will govern them. This means upskilling leaders in AI literacy, risk awareness, and ethical frameworks to ensure these systems are used wisely. As AI agents move from novelty to necessity, the Responsible AI conversation needs to shift from theory to application. Governance, training, and culture must evolve in step with capability. #ResponsibleAi #AI_Literacy #AI_Leadership

A Harvard study finds that leadership skills show up when managing GPT-4o agents, just like with real teams. The strong correlation (r=0.81) suggests that effective leadership is not just about authority, but about how we interact, even with AI: asking questions, active listening and shared turn-taking. Interesting how the same interpersonal skills that build human trust and performance also drive better outcomes from AI agents - social intelligence really is the new technical skill. Thanks Ethan Mollick #ResponsibleAI

How to Scale a HealthTech Startup Using Responsible AI. What is the biggest reason HealthTech startups fail so early? They’re using a traditional software services sales playbook in a emotionally charged sector that breathes complexity, regulation, and human impact. - You’re not selling convenience. You’re redesigning care delivery. - You’re not optimizing for conversions. You’re selling trust and outcomes inside workflows. Here’s where the traditional model breaks down and what you need instead: Responsible AI as strategy, not just compliance Clinical validation + explainability baked into the workflow Governance frameworks aligned with regulation and patient trust Interdisciplinary teams where domain experts shape the AI roadmap The stakes are real: 80% of AI projects fail - twice the rate of traditional IT (RAND, 2024) 123 major AI incidents were reported in 2023 - a 32% rise (Stanford HAI, 2024) 74% of companies paused at least one AI project due to risk concerns (Accenture, 2025) 60%+ of S&P 500 firms now cite AI as a material risk (Harvard Law School, 2024) The future of HealthTech isn’t just “AI-powered.” It’s responsibly AI-powered. 📘 Ready to lead with confidence? Our "AI Literacy for Leaders" course on Maven gives you the strategy, structure, and language to start scaling AI systems with trust and impact. #AI #Ecosystem #Innovation #Education #Accelerator #ResponsibleAI #Orcawise #RESPECT

🧭 Responsible AI: From Global Standards to Real-World Implementation. For years, we've debated what "trustworthy AI" should look like. Now, the shift is clear: global frameworks are converging—and implementation is next. This draft report by Robert Kilian, Linda Jäck, Dominik Ebel*aims to influence future legislative decisions and shape the EU's digital strategy. The EU AI Act, ISO/IEC standards, NIST RMF, and OECD principles are no longer abstract—they’re setting the baseline for what Responsible AI must deliver: transparency, accountability, bias mitigation, and societal well-being. Takeaway: Responsible AI is moving from theory to deployment. The paper outlines how a common foundation of international standards—from ethics to risk management—is shaping AI governance across sectors and jurisdictions. Insight - The next challenge is coordination: - Across borders - Across regulatory regimes - Across technical and ethical domains 📌 What’s coming next? Governments and industries will co-develop Technical Requirements + Implementation Mechanisms. These will define how to operationalize values like fairness, safety, and human oversight in AI systems—through Quality Management Systems, Bias Audits, Conformity Assessments, and AI-specific cybersecurity protocols. ✅ Already happening: ISO/IEC 42001 for AI QMS NIST RMF for AI Risk Drafts on bias, explainability, robustness The bottom line? Compliance will no longer be a checkbox—it will be a competitive differentiator. Responsible AI leaders aren’t waiting for enforcement. They’re building with trust, governance, and human values baked in—not bolted on. #AI Literacy sets the foundation for Responsible AI and the coming global frameworks convergence? #ResponsibleAI #AIethics #EUAIACT #ISO42001 #Governance #AIlaw #TrustworthyAI #OrcaPulse #AILiteracy

🔍 Thinking About an Exit in 2025? Here's What Acquirers Are Really Looking For. The M&A game has changed. AI isn't just reshaping how deals get done—it’s redefining who gets acquired and at what valuation. Here are some insights I found in the McKinsey’s 2025 M&A Report: 📌 A European industrials firm used GenAI to screen 500+ targets—prioritizing companies with “automation-native” operations and strong AI foundations. 📌 Acquirers are placing a premium on AI readiness: mature data infrastructure, AI-literate leadership, and embedded AI workflows. 📌 Companies with responsible AI governance and risk oversight are viewed as lower integration risk and higher long-term value. 🚀 What entrepreneurs need to know: If you're preparing for an exit in the next 12–24 months, AI readiness is now part of your valuation narrative—whether you're in services, manufacturing, SaaS, healthcare, or beyond. ✅ Build internal AI capabilities—don’t just bolt them on. ✅ Formalize Responsible AI frameworks. ✅ Upskill leadership teams in AI literacy and governance. ✅ Document how AI supports real business outcomes, not just experiments. In 2025, the winners won’t just be AI-powered—they’ll be AI-proven. #AILiteracy #ResponsibleAI #MergersAndAcquisitions #ExitStrategy #AIReadiness #BoardLeadership #DigitalTransformation

Why Responsible AI Isn’t Optional Anymore—It’s the Differentiator. AI is scaling fast—but trust isn’t keeping up. The latest research shows a clear split between frontrunners and everyone else. 📊 74% of companies paused at least one AI project in the past year due to risk concerns. 📊 60%+ of S&P 500 firms now list AI as a material risk in their annual reports (Harvard Law, 2024). 📊 Companies with high RAI maturity see 18% higher AI-driven revenue (Accenture, 2025). 📊 AI-related incidents—bias, deepfakes, hallucinations—rose 32% in 2023 (Stanford HAI, 2024). 📊 48% of organizations globally reskilled staff in 2024 to meet changing tech demands (WEF, 2025). And yet—Responsible AI is still treated by many as a bolt-on, not a build-in. The winners? They're embedding RAI into everything: governance, culture, workflows, compliance, and talent strategy. ✅ They're reskilling their teams with AI literacy. ✅ They're aligning risk management with innovation. ✅ They understand: trust = scale. This is why we built AI Literacy for Leaders—to give executives, advisors, and change-makers the playbook to lead responsibly and move fast without breaking trust. 🚀 Enrolments open for April and May cohorts. 🔗 Link is in the comments #ResponsibleAI #AILiteracy #AILeadership #EUAIACT #Governance #Ethics #FutureOfWork

AI & Data Literacy: What Business Leaders Need to Know in 2025. As Responsible AI continues to move from principle to practice, the new 2025 State of Data & AI Literacy Report offers hard numbers, and a clear direction. Here are my top 10 insights derived from the report: 1. AI literacy is accelerating faster than data literacy. 69% of leaders say AI literacy is now essential to daily tasks (up 7% from last year), while data literacy remains flat at 86%. The focus is shifting, and fast. 2. The skills gap remains significant. 60% of leaders report an AI literacy gap in their organization. For data literacy, it's still 50%. Awareness isn't the issue. Execution is. 3. Literacy drives performance. AI-literate employees outperform, say 65% of leaders. The figure rises to 78% for those with strong data skills. Literacy is no longer a nice-to-have, it’s a competitive edge. 4. These skills command a premium. 71% of leaders are willing to pay more for candidates with AI skills. 12% would offer a salary bump of 40–50%. For data skills, 15% would pay 30–40% more. 5. Innovation and decisions go hand in hand. AI literacy enables innovation. Data literacy powers sound decisions. 37% say lack of AI skills slows innovation. 40% cite poor data skills as a risk to productivity. 6. Responsible AI isn’t optional. 74% of leaders highlight Responsible AI ---- ethics, bias mitigation, safety, as a core skillset. This is where technical literacy meets strategic leadership. 7. Modern AI agents are built to reason, not just respond. New agent-based systems use ReAct, Chain-of-Thought, and Tree-of-Thought to plan, evaluate, and adapt. These are the building blocks of AI that thinks before it acts​. (Source: Agents Whitepaper, p. 10) 8. Adoption is high, but not yet widespread across functions. 64% of teams are actively using AI. 82% use it weekly. Yet only 34% of marketing, finance, and ops teams have adopted AI tools, an obvious gap, and a growth opportunity. 9. Upskilling delivers measurable ROI. Organizations with mature literacy programs saw: → 95% faster decision-making → 92% revenue increases → 91% cost reductions This is not about potential, it’s already happening. 10. Blended learning is winning. 42% of leaders rely on blended learning, online and live, to build skills at scale. Interactivity and role relevance matter more than content volume. Source of Insights: 2025 State of Data & AI Literacy Report & Agents White paper 2024

Global Healthcare AI Regulation: A Leadership Test for the U.S. Despite widespread agreement among industrialized nations that AI in healthcare must be regulated, meaningful international collaboration remains elusive—primarily due to political fragmentation, especially in the U.S. As the incoming Trump administration reconsiders the direction of U.S. AI policy, the future of President Biden’s executive order remains unclear—creating uncertainty around America’s global AI leadership. While pre-generative AI is being managed under existing medical device frameworks in the U.S., U.K., and Canada, no country has yet established enforceable rules for generative AI in healthcare. The EU is furthest ahead with its AI Act, classifying most clinical AI applications in healthcare as high-risk. Japan’s agile governance model offers a more flexible, stakeholder-driven approach. Prediction If the U.S. continues to stall, the EU and other proactive nations will set the de facto global standards for healthcare AI—potentially sidelining American companies from key international markets. Takeaway To stay globally competitive and build trust in AI-driven healthcare, countries—and especially the U.S.—must shift from fragmented national strategies to coordinated international frameworks. The window for leadership is closing. Those who act boldly now will shape the future.

MIT Technology Review Insights, “A playbook for crafting AI strategy.” In the race to adopt AI, many businesses are discovering that “responsible” deployment is just as critical as achieving quick wins. Here are the top insights: • Nearly every organization (98 percent) would rather slow down AI projects than risk rolling out an unsafe or unsecure system. That’s a clear sign that trust, transparency, and robust governance are non-negotiable. • Governance, security, and privacy are the biggest brakes on AI adoption, flagged by almost half (45 percent) of surveyed companies. Executives recognize that poorly governed AI isn’t just a technology issue, it’s a serious reputational and regulatory risk. • “Hallucinations” and faulty outputs can lead to real damage if unchecked. Businesses must implement thorough validation steps, bias audits, and oversight to keep models in line with both ethical standards and legal requirements. • Data privacy and protection are top of mind as regulation tightens worldwide. Organizations are embracing risk-based frameworks to classify AI tools as low or high risk and implementing audits to stay ahead of new rules. • The responsible approach to AI goes hand in hand with strong data quality. A model is only as good as the data it sees. Ensuring accurate, bias-free data and storing it securely which ultimately results in better outcomes and fewer ethical pitfalls. Leaders who balance innovation with a well-grounded commitment to safety, privacy, and integrity will position their organizations, and the AI industry, on the strongest possible footing.

AI Agents are here now, not in 10-years. AI Agents Are Transforming Decision-Making: Embracing Responsible AI Governance. Thanks to Jam Kraprayoon and his colleagues at the Institute for AI Policy and Strategy (IAPS) for AI Agent Governance: a field study. In the era of autonomous agents making decisions is no longer a distant future; it's our present reality. Companies like Klarna and Google are already leveraging AI agents in customer service and code generation, marking a significant shift in how tasks are accomplished. However, despite their potential the reliability of these agents remains a pressing concern. Issues such as struggling with intricate tasks, hallucinations, looping behaviors, or silent failures pose significant risks, especially in critical systems where such malfunctions can have severe consequences. The challenges extend beyond technical malfunctions to encompass broader societal implications. From the possibilities of malicious exploitation and loss of control to the far-reaching impacts on jobs, inequality, and power dynamics, the deployment of AI agents demands a nuanced approach to governance. Responsible AI transcends mere considerations of fairness and transparency; it necessitates robust governance mechanisms across various dimensions: - Alignment: Are these agents truly aligned with human interests? - Control: Can we intervene and deactivate them when necessary? - Visibility: Is it possible to track and audit their decision-making processes? - Security: Are these agents resilient against cyber threats and attacks? - Societal Integration: Do they promote fairness, equity, and overall accountability? The key takeaway is clear: designing efficient AI agents is just the first step. Establishing scalable governance frameworks is imperative. This involves crafting regulations, developing tools, setting standards, and intriguingly, utilizing agents to assist in governing other agents. While the field of Responsible AI is still evolving, the implications are profound. The time has come to shift focus from mere speculation to building the necessary infrastructure to govern AI agents effectively.

🔥 From Margins to Momentum: AI Literacy in the Global South. The CFTE AI Literacy Whitepaper (v0.3 – 2025) delivers a powerful message: AI literacy is no longer optional—it’s foundational. Not just for engineers, but for entire societies. Some key takeaways: 🌍 Access-first, community-driven learning In Rwanda, “Digital Ambassadors” bring AI education to remote areas. In India, over 50,000 rural students are learning to build chatbots and explore AI ethics. 🤝 Responsible AI isn’t a luxury—it’s a necessity Brazil’s national dialogue on AI ethics shows how the Global South is embedding cultural values, accountability, and collective governance into their AI strategies. 🧠 AI Literacy = More than technical skills It's about trust, empowerment, and using AI to solve local challenges—from crop disease detection in South Africa to Bahasa-based edtech models in Indonesia. 🗣 Language = Digital Sovereignty Local LLMs in native languages are preserving culture and asserting agency—ensuring oral traditions and indigenous wisdom aren’t lost to globalized AI systems. 👩🏽‍💻 Equity at the center From youth to civil servants, from women to informal workers—programs like Kenya’s Ajira Digital prove inclusive AI literacy can scale when governments, academia, and civil society unite. 🔑 The shift: From passive users to active innovators This is more than catching up. The Global South is showing how AI literacy, grounded in context, ethics, and community, can ignite homegrown innovation. 📣 If Responsible AI is the goal, AI Literacy is the starting point—globally. Source: The CFTE AI Literacy Whitepaper (v0.3 – 2025)