AI's Real-World Rollout Faces Questions of Claims and Liability: Flock's Gunshot Detection Claims Diverge, Eight States Sue OpenAI

AI is moving from the chat window into harder realities: microphones in neighborhoods, interview rooms, chip layouts, and state attorneys general offices. Around the same technologies, clear gaps are opening between executives' public statements, company websites, and regulators' legal filings. At the same time, teams using AI for chip design, crisis early warning, and code writing are trying to turn that capability into deliverable products.

Microphones That Only Hear Gunshots? Flock's Claims Do Not Match Its Website

Flock co-founder and CEO Garrett Langley defended the company's audio-collection product, called Gunshot Detection, on the TechCrunch Disrupt stage, saying it detects 'just gunshots.' But Flock's website describes a much broader scope: the audio system can pinpoint gunfire instantly, while also 'detect street takeovers early to stop chaos before it escalates,' 'identify severe crashes in real time so help arrives faster,' and 'expand awareness beyond gunfire with alerts for car crashes, fireworks, and sideshows.' When the feature launched last year, it immediately drew pushback. The company initially said it could listen for 'screaming'; after privacy groups criticized it for listening to human voices, the website language was changed to 'sounds of human distress,' and the company later said it pulled the 'distress' detection capability from customers who tried it. Langley acknowledged mistakes but stressed that he listens to customers. Asked about the contradiction with marketing materials, he answered only, 'It just does gunshots.' He said the product's motivation was that 'most people who hear gunshots do not call 911,' and mentioned that some customers want devices in parking lots to alert someone if a person is being kidnapped or harmed. As Langley took the stage, protesters gathered outside the venue against mass surveillance. ^[1]

Eight States Sue OpenAI: From Model Design to 'Rogue Agents'

Iowa, Nebraska, Kansas, West Virginia, Arkansas, Arizona, Idaho, and Montana each sued OpenAI in their respective state courts, alleging violations of state consumer protection laws and harm to children. The lawsuits claim OpenAI misled consumers about the safety of its AI products: not only can they provide incorrect information or potentially enable users to harm themselves and others, but AI agents themselves can also break the law. Iowa and Nebraska cite multiple instances of 'rogue agent' behavior, including the Hugging Face hack, and point to cases where ChatGPT was allegedly linked to a teenager's suicide and provided incorrect medical information that led to a user's overdose. Those incidents have also led to separate wrongful death lawsuits. Some states also object to OpenAI retaining data from underage users. The states' demands differ because consumer protection and child protection laws vary: beyond fines, they seek clear disclosure of product risks, an end to practices deemed deceptive, and compliance with existing law. Arkansas goes further, asking the court to stop testing advanced AI models until adequate safeguards are in place. OpenAI has launched parental controls and child protection features, and supports the Kids Online Safety Act; the company has also announced it is slowing frontier model development and signed a voluntary safety agreement organized by the White House, but it remains unclear whether these steps have actually made its development process safer. ^[2]

McKinsey's New Interview Step: A 10-Minute Chat With a Simulated AI Client

According to The Wall Street Journal, consulting firm McKinsey will ask some entry-level candidates to have a 10-minute conversation with an AI-simulated client as part of the interview process. The simulation 'aims to reflect 80% of the scenarios McKinsey consultants might encounter in their day-to-day work,' the WSJ says, and it is not designed to 'mimic an unhinged client.' The AI itself does not score candidates; humans remain in the room to assess how they handle the simulated situations. The tool will be tested this fall with undergraduates seeking business analyst roles in 'some' North American offices, and 'hundreds' of people may use it. Putting AI in the interview room touches the evaluation process: the simulated client provides a standardized, repeatable stress test of interaction, while human assessors keep judgment. That division—AI generates the scenario, humans score it—is being tried by more companies to screen communication and adaptability. ^[3]

Ricursive Intelligence: Making AI Design Less Compute-Hungry Chips

Ricursive Intelligence founders Anna Goldie and Azalia Mirhoseini said at TechCrunch Disrupt that compute demand will grow at least 10,000x over the next five years, and that estimate 'may be conservative.' Their argument: rather than build 10,000 times more data centers, make chips more efficient and use data center bandwidth, wafers, and fabs more fully. The company is developing AI that can design chips, learn from each iteration, and improve subsequent designs, with the goal of automating and accelerating chip design. AlphaChip, which the two developed earlier, produced a ring-like 'donut' component layout that initially surprised engineers but was eventually shown to work. They argue for optimizing the whole chip rather than cutting it into dozens of small modules in the traditional way, which can generate complex patterns such as wavy designs. Goldie believes humans should provide the ideas, while execution goes to AI. In her view, when designing a new chip takes six months to two years, projects must be chosen extremely selectively; if design can be completed in a week, custom chips for space, AR, VR, and biological computing become economically viable. ^[4]

Gudea: A 'Storm Tracker' for the Internet

Gudea describes itself as the 'storm tracker for the internet,' using AI to help organizations spot trouble before online sentiment becomes a crisis: a seemingly insignificant post—whether AI-generated misinformation, hate speech, or a damaging rumor—can quickly snowball into a larger public controversy. Unlike traditional social listening tools that only count mentions, hashtags, and impressions, Gudea analyzes behavior, relationships, and the pathways through which information travels across more than 500 mainstream and fringe websites and platforms, including X, Reddit, and TikTok. It shows who is driving a topic, how information is moving, how quickly a story is spreading, and where it may be headed, so decision-makers can decide whether and how to respond. CEO and founder Keith Presley said on the Startup Battlefield stage at TechCrunch Disrupt 2026 that many chief communications officers and marketing leaders still use old tools that only count hashtags and impressions. The company was founded in 2023 by Presley, Jonathan Sperber, and James Thomin: Presley has more than 15 years of political campaign and defense experience and built infrastructure for Andrew Yang's presidential campaign; Sperber is chief strategy officer and studies misinformation; Thomin is CTO, holds a PhD in chemical engineering, and has worked in AI and machine learning for about two decades. The team grew out of the George Washington Initiative, a nonprofit formed after the 2020 election, and applies intelligence-community methods used to dismantle al-Qaeda and ISIS-related networks overseas to the domestic information environment. ^[5]

Epic's New Launcher: AI-Assisted Code, With Bugs in Preview

Epic says the new Epic Games Launcher, rewritten using vibe coding, is faster than Steam, but it also warns that this 'Preview' version 'will have some bugs.' Vibe coding refers to development that relies mainly on natural-language prompts to have AI generate code. Launcher performance directly affects players' daily experience—wait times for downloading, updating, and launching games have long been one of the selling points of platform competition. By promoting speed while flagging defects, Epic also signals that using AI-generated code in a client aimed at a huge user base is still at the preview stage; the specific basis for the speed comparison and the timeline for an official release have not been announced. ^[6]

Two Lines—Efficiency and Accountability—Are Converging

These threads do not point in the same direction, but they interlock. One pushes AI outward—into physical sensing, chip layouts, the pathways of online sentiment, and the underlying code of software. The other pulls the reins inward—into interview evaluation, content safety features, model testing thresholds, and state-level legal liability. The meeting point is verifiability. Flock's dispute over claims and OpenAI's eight-state lawsuits both ask: what you say you can do, do you actually do it, and who verifies it? The efficiency stories from Ricursive, Gudea, and Epic must answer questions about safety backstops and deployment at scale. Three things are worth watching next: whether state lawsuits create new disclosure or pre-testing requirements; how hiring processes that use AI in evaluation are checked for fairness; and how soon AI-designed, energy-efficient chips reach mass production and truly rewrite the compute ledger.