Lance Eliot

The FTC has unveiled a new AI policy proposal targeting generative AI and large language models. This initiative aims to compel AI makers to disclose biases that lead to less-than-truthful outputs, aligning with the FTC Act's consumer protection mandate. The agency asserts that consumers reasonably expect AI to provide accurate information, and any deviation without clear disclosure constitutes deceptive practice. While not banning bias, the proposal requires AI companies to inform users if their systems prioritize objectives different from what users expect. This follows previous FTC efforts against false AI advertising. The proposal, issued July 1, 2026, is open for public feedback until July 31, 2026, sparking debate over potential overreach versus necessary consumer safeguards. Defining "truthful" and effective disclosure methods remains a complex challenge.

Google has unveiled a new AI governance framework, proposing an independent Frontier AI Regulatory Organization (FARO) for the U.S. This initiative aims to fill the current void in federal oversight for leading-edge AI, such as large language models, which currently operate in a "Wild West" scenario. The proposal seeks a balanced approach, avoiding both excessive regulation that could hinder innovation and insufficient controls that might lead to existential risks. FARO would be tasked with establishing standards, mitigating risks, and ensuring transparency for frontier AI. However, the plan sparks debate over its "middle ground" assertion, its narrow focus on frontier AI, and the practicality of a new independent body versus existing government agencies. Concerns include potential regulatory capture, slow adaptation, and neglecting risks from other AI types. The article underscores the critical need for ongoing discussion to effectively manage AI's rapid advancements.

AI developers are grappling with a "jurisdictional morass" as state-level AI laws proliferate, creating a complex and often contradictory regulatory environment. Major generative AI models, designed for universal application, struggle to meet these disparate legal requirements. Without a unifying federal law, AI makers must painstakingly reconfigure their systems to comply with each state's unique stipulations, risking substantial penalties for non-compliance. This challenge transcends typical software localization, demanding fundamental alterations to AI's core reasoning and output. The future necessitates AI architectures built with inherent, flexible jurisdictional compliance, meaning user experiences will increasingly diverge based on location and local legal frameworks. This marks a profound shift in AI's operational paradigm.

AI policy is now a universal concern for all lawmakers, especially those in Congress. A complex legislative landscape sees numerous state AI laws emerging, while a unifying federal framework remains elusive, sparking debates on balancing innovation with societal protection. Crucial questions include federally regulating frontier AI models, restricting AI companionship, addressing federal preemption of state laws, limiting AI makers' liability, and protecting workers from AI job losses. Lawmakers must enhance their AI policy literacy to judiciously navigate these pressing challenges and suitably shape the future of our country.

Anthropic researchers have unveiled a significant discovery within large language models (LLMs): an internal "scratchpad" or "working storage" dubbed "J-space." Using a new "Jacobian lens" (J-lens) technique, they found this area holds "verbalizable representations" that critically influence an LLM's output. An experiment demonstrated that altering J-space content, like changing "spider" to "ant" when asking about a web-spinning animal's legs, directly changed the AI's answer from eight to six. This breakthrough offers insights into LLM mechanics, potentially improving them but also raising security concerns. While Anthropic draws parallels to human consciousness's "global workspace theory," experts and the company itself urge caution, stressing that it's premature to claim AI consciousness based on this finding. The debate continues on what truly constitutes AI consciousness.

Generative AI and large language models are increasingly providing "therapy micro-bursts," offering immediate, on-demand mental health advice to millions. Users access AI 24/7 for quick psychological assistance, often in short, variable durations, a stark contrast to traditional structured therapy. The article outlines seven types of these micro-bursts, including emotional regulation, cognitive reframing, and decision support. While offering accessibility and rapid cognitive offloading, significant risks exist. Concerns include AI dispensing unsuitable advice, fostering delusions, and lacking robust safeguards, which has already led to lawsuits against AI makers. This dual-use technology presents both immense upsides for mental health support and considerable downsides, necessitating careful management to mitigate harm and maximize benefits in this global experiment.

Hackers switch their prompts to less common natural languages to avoid AI safety features. AI makers are coping with it

AI laws, like the EU AI Act, mandate chatbots disclose their artificial nature to users, starting on August 2, 2026, with some US states adopting similar rules already. This aims for transparency, preventing confusion between human and AI interactions. However, the efficacy of these disclosure laws is widely debated. Critics question if users are genuinely unaware or if such legislation is more performative than substantive. A key concern in the EU Act is its "reasonable person" clause, which exempts disclosure if AI interaction is "obvious," creating significant ambiguity. This vagueness, also arising in some US state laws, raises questions about potential loopholes and whether these minimalist regulations will truly protect users.

New AI laws are creating a chaotic regulatory landscape, largely due to their imprecise nature and lawmakers' limited understanding of artificial intelligence. This vagueness forces AI makers into a precarious position, as they must interpret ambiguous regulations to ensure compliance. Without a comprehensive federal framework, individual states are enacting diverse and often conflicting statutes, leading to a "jurisdictional morass." Consequently, different AI companies may interpret the same law differently, resulting in inconsistent AI behaviors and potential user confusion. The article emphasizes the critical need for lawmakers to prioritize crafting more precise and unambiguous AI legislation to foster effective governance and prevent a fragmented, unpredictable regulatory environment.

The article underscores the critical importance of AI governance within the public sector, distinguishing it from private sector applications by its focus on democratic values and the rule of law. Many government entities either hastily adopt AI or unknowingly integrate it, underscoring the need for a robust framework. A new book, "Governing With AI," is highlighted as a vital guide for public leaders navigating this complex landscape. The public sector faces unique challenges, including limited resources, a lack of specialized expertise, and political complexities. Given the high stakes of AI in public services, which directly impact citizens' lives, a higher standard of governance is imperative. The ultimate goal is to balance AI's potential benefits, like efficiency and fraud detection, with comprehensive oversight to prevent adverse outcomes.