Leadership is asking: are we getting value from AI? Which tools are worth the spend? Where are we exposed? Right now, most teams have no idea.

Harmonic Security Usage Explorer changes that.

You get a complete picture of how your organization uses AI, automatically categorized into custom tasks and use cases.

You’ll see the projects being worked on, who’s using what tools, where AI investments are driving value, and where employees are engaging in risky behavior.

CIOs can rationalize spending and cut wasted licenses. CISOs can pinpoint where risk exists and neutralize it. AI committees can show exactly how their efforts are paying off.

See it in action

Kids Outlearn AI, and Nobody Knows Why

A report from MIT Technology Review puts a sharp point on one of cognitive science's most embarrassing open questions: human children acquire fluent language from roughly 10 million words of input by age 10, while large language models require exposure to hundreds of billions of tokens to achieve comparable performance. The disparity is not marginal. Children also generalize from fewer examples, learn from social context rather than raw text, and self-correct without gradient descent. Researchers have proposed several candidate explanations, including embodiment (the social scaffolding of caregivers), and innate grammatical structure, but no single account has closed the gap. The honest answer is that we do not know how young humans learn as they do, which means every claim that AI has "solved" language should carry a healthy dose of skepticism.  In my opinion, this also lends further credence to alternative modes of constructing intelligence, such as Yann LeCun’s Joint Embedding Predictive Architecture (JEPA).  Dr. LeCun argues that so-called “world models” are a better conduit to human-level machine intelligence than LLMs.

The Consciousness Debate Can Be a Distraction.

There is an ongoing public argument over whether AI systems are conscious or sentient. This is not merely philosophically slippery; it is actively warping how we allocate research funding, draft regulation, and assign moral weight in policy discussions. A recent piece in MIT Technology Review argues instead, that the framing itself is the problem, pulling attention toward the unprovable while systems with real, measurable effects on employment, autonomy, and information quality operate without adequate scrutiny or governance. When the debate is whether an AI has feelings, the question of whether an algorithmic hiring system is discriminatory (a more pressing issue) tends to get less airtime.  It’s also worth nothing that an algorithmic hiring system may reflect the bias of the humans who created it. Precision about what these systems actually do, statistically and mechanistically (i.e., their impact), is more meaningful for us than any verdict on their inner life.

How AI is Transforming Intellectual Attribution

As AI moves from a supporting tool to a named contributor in pharmaceutical discovery, intellectual property law is straining under the weight of what amounts to a paradigm shift. The current legal framework in most jurisdictions requires a human inventor, but in several recent drug development cases, the AI's generative contribution to molecular design was substantive enough that listing a human as sole inventor has started to feel like an exaggeration. Scientists, patent attorneys, and bioethicists are now actively debating three distinct questions: what counts as intellectual contribution, whether institutions or companies rather than individuals should hold AI-assisted patents, and how credit structures affect researcher incentives. The answers will shape who profits from the next generation of therapeutics, and the legal infrastructure is not keeping pace with AI-enabled science.

OpenAI Handed Mathematicians an Existential Crisis

The Verge reports that OpenAI's recent solutions to longstanding open problems in mathematics have sent a detectable tremor through the professional mathematics community. I recently watched an interview with famed mathematician Terence Tao saying that even high schoolers may soon be able to contribute to the field, with the aid of AI. The concern is not that the proofs are wrong. It is that they appear to be right, and they were produced by a system that does not understand, in any recognizable sense, the actual purpose of the proof it creates. Leading mathematicians interviewed for the piece describe a genuine crisis of purpose: if a machine can close problems that occupied careers, what is the social and intellectual value of mathematical expertise? Some researchers draw a distinction between verification and discovery, arguing that knowing which questions to ask and explore still remains irreducibly human. Others are less certain. The debate is nascent and the anxiety is real.

AI Training and Copyrighted Books: Still No Clean Answer

TechCrunch lays out the current legal landscape around AI training data and copyrighted text, and the headline summary is that no definitive ruling exists. Several active lawsuits from authors and publishers are working through U.S. courts, with “fair use” as the central contested doctrine. The core argument from AI developers is that the process of model training is in and of itself transformative, and does not reproduce the original work in any commercially harmful way.  The counterargument is that the economic value of an author's text is being captured without compensation or consent. Two complicating factors are that training datasets are rarely disclosed in full (remember the continuing polemic around closed-source models), and that courts have yet to rule in a case with facts squarely on point. Until they do, the legal status of the raw material underlying most major AI systems remains genuinely unresolved.

Uber Faces a Nearly $1 Billion Fine Over Algorithmic Discipline

European regulators have levied a fine of nearly $1 billion against Uber under GDPR, according to TechCrunch, specifically targeting the company's use of automated systems to suspend and penalize drivers without meaningful human review.  This is a governance issue in real time.  The enforcement action centers on the principle that consequential decisions affecting individuals' profession cannot be delegated entirely to algorithms without transparency, explanation, and a right to contest outcomes. This is not the first GDPR fine involving automated decision-making, but the scale signals that regulators are willing to use the full force of existing law rather than waiting for new AI-specific legislation. For every company using algorithmic systems to manage workers, this ruling is a concrete liability benchmark, worthy of watching for those interested in AI governance.

OpenAI Launches a Blog About Power. Make of That What You Will.

OpenAI has introduced AI Futures, a dedicated publication focused on AI's effects on governance, individual freedom, and the distribution of power, as reported on OpenAI's own site. The framing is deliberately civic: the blog positions itself as a space for serious engagement with questions about who controls powerful AI systems and what that means for democracy and human agency. The tension is obvious. A company that controls some of the world's most capable AI systems is now publishing analysis about the dangers of AI power concentration. That does not automatically make the content wrong, and some of the questions it proposes to engage with are exactly the right ones. It does mean readers should bring their own views to whatever conclusions the publication reaches. It’s difficult for an industry participant to regulate their own work.

Worth Your Time

Talking to Strangers by Malcolm Gladwell is not about AI, but it may be the most useful book you can read right now for thinking about automated judgment. Gladwell builds a detailed case around the psychology of how humans assess one another, drawing on cases including the bail-setting algorithm research by Sendhil Mullainathan and colleagues, which found that a machine trained on historical data outperformed judges in predicting flight risk. The book is flawed in places, and some of the science has been questioned, but the core provocation holds: human intuition about other humans is less reliable than we assume, and the question of when to defer to a model versus a person is one we are answering by default rather than by design.  However, Gladwell’s work is worth viewing in context, as humanity confronts new challenges around algorithmic decision-making.

The human mind is the original generative machine.