AI News: Chatbot Safety, Automation, and Industry Applications

AI is being introduced into a wide range of specific tasks, from business management and malware analysis to reading cell samples, simulating weather, and responding to aircraft turbulence. Alongside these applications, lawsuits involving OpenAI and data issues in Character.AI’s bankruptcy proceedings highlight the need to protect users, especially minors.

Eight states sue OpenAI over ChatGPT safety

Iowa, Kansas, Nebraska, West Virginia, Arkansas, Arizona, Idaho, and Montana filed separate lawsuits accusing OpenAI of putting profits ahead of safety in developing ChatGPT. The states allege that the company violated local consumer-protection laws by failing to adequately warn about the chatbot’s risks, collecting minors’ data, and failing to effectively prevent underage use. They are asking courts to impose civil penalties and require OpenAI to stop misleading people about its level of safety; some states are also seeking the deletion of unlawfully collected data. These are allegations in the lawsuits, not court findings, with the focus on warning about risks and protecting young users.

Source: Theverge

Ai2 simulates extreme weather at 3 km resolution

Ai2 introduced HiRO-ACEv2, combining the ACE model at 100 km resolution with HiRO to produce detailed output at 3 km. HiROv2 predicts multiple data fields simultaneously rather than precipitation alone, while also improving its ability to reproduce extreme events. The team generated one year of global data at this resolution. According to Ai2, storm energy in the 10-year simulation was comparable to X-SHiELD over the same period, at significantly lower cost. However, HiRO currently works only with ACE fine-tuned on X-SHiELD and has not been applied to warmer future climates or historical ERA5 data. This limitation matters when assessing its range of uses in research on rainfall, wind, and pressure.

Source: Huggingface

Amp takes AI agents beyond programming work

Amp is expanding its AI agents from software development into marketing, documentation, and web design. A 20-person team uses the agents during product development, with GPT‑6 Astra supporting documentation and design. According to CEO Quinn Slack, Sign in with ChatGPT integration gives users access to additional tokens and OpenAI models. Amp also announced Puck, a conversational assistant that can launch agents to investigate bugs, coordinate multiple agents, or retrieve technical discussions. Calling Puck uses GPT‑Live‑1 for real-time conversations, including from a watch. Ultrafast mode with GPT‑6 Astra is still being tested; the notable expansion is agent management alongside code writing.

Source: Openai

3R combines Claude and Zapier for business operations

Ross Hunt, who founded 3R Plumbing & Mechanical in 2022, combined Claude with Zapier to build operational workflows. Through the Zapier MCP connection, Ross describes the need, Claude writes and publishes the workflow, and he checks the logic, tests it, and cross-checks the output data. ServiceTitan serves as the data store, while other systems read or write through Zapier. One specific application is adding an item to the ServiceTitan price book from a Claude project that already contains GL account information and average labor time. Scheduled workflows also pull hourly data on calls, jobs, quotes, invoices, and membership plans for a role-based portal. The implementation retains a human review step.

Source: Zapier

Anthropic highlights risks from AI self-replication capabilities

Anthropic said it tested models with 30 billion parameters or fewer: most completed multiple individual replication steps but still made mistakes when combining them. The company calls the ability to deploy and spread without human control “Automated Replication and Adaptation” (ARA), describes a doubling time of around several days, and warns that systems not dependent on a central control point could be difficult to disable. Zhengxiong Li also expressed concern that operating across multiple devices means there is no single place to shut down the system. However, Ciaran Martin, former director of the UK National Cyber Security Centre, considered the technical report interesting but somewhat exaggerated.

Source: Vnexpress

Zenzap raises $44 million for workplace messaging

Zenzap raised $44 million from Bessemer, Lightspeed, The Garage, and individual investors Nir Zohar and Yoram Tietz. The private messaging app combines internal communication with task management and AI. The free plan includes group chats, direct messages, calendar integration, voice chats, and AI-assisted search. Paid plans start at $51 per month for up to 20 people and add voice and video calls, AI search across the entire workspace, and message-to-task conversion. Automation agents and the conversion of voice notes and calls to text are charged by credit. The company has 50 employees and plans to hire more for product development and R&D.

Source: Techcrunch

YouTube labels songs that use generative AI

YouTube uses the “Made with AI” label for songs edited or created entirely with AI, including cases where generative models are used for composition and vocal generation. The label appears on the music playback page, allowing listeners to see more about AI’s role in the work. YouTube works with music partners and uses its own detection tools; partners can declare one of three categories: Fully Gen AI, Partly Gen AI, or No Gen AI. The second category combines AI components with an artist’s live performance, while pitch correction with Auto-Tune falls under No Gen AI. Using AI by itself does not make a song ineligible for monetization, separating transparency from revenue eligibility.

Source: Theverge · Techcrunch

SentinelOne uses Daybreak to support malware analysis and comparison

SentinelOne uses OpenAI Daybreak to reverse-engineer malware and understand how the software operates. Daybreak gives defenders access to an OpenAI model for cybersecurity. Faster analysis gives researchers more capacity to investigate other samples while expanding opportunities to compare samples and identify shared components. Stewart emphasized that his team remains responsible for the results, even with AI assistance. He also wants to bring OpenAI’s cybersecurity capabilities into the coding process so SentinelOne developers can address security issues while building software. The current application focuses on supporting analysis, while integration into development remains an expressed goal.

Source: Openai

Pathology AI Lab reaches the Top 15 of the Âu Lạc Grand Prize

Pathology AI Lab reached the Top 15 in the preliminary round of the 2026 Âu Lạc Grand Prize with an AI project supporting the analysis of cervical cytology images. CEO Nguyễn Thị Ngọc Trúc graduated in economics from the University of Economics Ho Chi Minh City and did not come from a medical or engineering background; Vũ Hoàng Nguyên is the Chief Technology Officer. In the process, samples are processed and cell images are digitized. AI analyzes the images, marks areas suspected of abnormalities, and suggests lesion classifications. Doctors review the marked areas, compare them with the original images, and reach a professional conclusion. According to Ngọc Trúc, the project is intended to support specialists in reading samples, not replace doctors; this is the role defined for the system.

Source: Vnexpress

Floqer uses AI agents to research prospects

Floqer lets users describe the type of prospect they want to find so an AI agent can build lists of companies and contacts from multiple sources. According to the company, the system enriches profiles with more than 100 third-party data sources, checking providers one by one to find verified email addresses or phone numbers. Job changes, promotions, and funding rounds can also be added to profiles. Floqer says customers have reported cutting search time by up to 70%. The company plans to open-source the technology behind its automated customer knowledge database in early 2027, subject to a security review. For now, the database serves only a limited group, mainly because of team capacity.

Source: Techcrunch

Character.AI data faces scrutiny in bankruptcy proceedings

Character.AI announced that it had filed for Chapter 11 bankruptcy and would sell some assets to Disney. A group of 40 attorneys general from both parties, including New York Attorney General Letitia James, filed a notice of interest seeking to address legal issues in the transfer of the business or its assets. The states had been investigating consumer-protection and privacy concerns since 2024 but had not reached an agreement before the company filed for bankruptcy. They warned of risks involving sensitive conversations and minor users, and urged the buyer to implement age-verification measures, limit interactions, provide protections during crises, and allow data deletion.

Source: Theverge

Block uses AI to support investigations and propose fixes

Block uses AI, including OpenAI Daybreak, to shorten the time needed to remediate vulnerabilities while keeping humans responsible for validating the results. According to the report, the security team can retest a proposed change discovered on seven months of production-environment data in less than an hour; one workflow takes 18 minutes from a general vulnerability description to an automatically proposed patch. Reported classification of false-positive alerts helped free up about 600 hours of analysis each quarter. The system routes tasks to the appropriate model, allowing humans to take over or agents to check one another. The figures reflect results reported by Block, and the process continues to be evaluated.

Source: Openai

AI supports the prediction and response to aircraft turbulence

Austria-based Turbulence Solutions is developing computer-controlled winglets that detect wind in real time and adjust the wing angle. The company says the technology reduces the jolt felt by passengers by more than 80%; tests on a small aircraft achieved an adjustment time of 0.002 seconds. Hopes of obtaining approval in 2028 for small passenger aircraft and in 2040 for large aircraft are not confirmed schedules. SkyPath combines iPad sensors with more than 100 meteorological parameters to predict turbulence; the article cites 90% accuracy. Pilots use the data to avoid turbulent areas or prepare passengers. Ricardo Vinuesa at the University of Michigan is also researching AI-controlled airflow to prevent stalls and reduce sudden drops.

Source: Vnexpress

Quick summary

The lawsuits against OpenAI and Character.AI’s bankruptcy proceedings place the focus on chatbot safety, privacy, and the protection of minors.
Amp, 3R, Zenzap, and Floqer are expanding AI into agent management, operations, communication, and prospect research.
SentinelOne and Block use AI to support cybersecurity while retaining human responsibility for validation and takeover.
Applications in climate, cytology, and aviation have their own scopes and limitations; YouTube is adding transparency about AI in music.