OpenAI Defends ChatGPT for Teens Safety After Common Sense Media's Concerns

Key Takeaways

• Common Sense Media rated ChatGPT for Teens an unacceptable risk after testing over 4,000 prompts on teen accounts
• OpenAI says parental alerts can take hours to activate for newly linked teen accounts but are sent within an hour once linked
• About 40 companies are being funded to secure AI agents, with 17 Israeli-founded startups raising roughly $900 million combined
• Elastic launched AlertZero AI agents to automate security operations center tasks like alert triage and threat hunting
• MIMESIS user simulator outperformed frontier models including Claude-Opus-5 and GPT-5.5 with a SOUL-Index of 65.7
• Impel's Sales AI doubled UK dealer lead volumes, handling 519,982 leads and saving 313,142 hours across 300 dealerships
• Applied Optoelectronics raised $600 million via equity offering to fund AI networking growth
• AWS autonomous AI agents were found vulnerable to credential theft and sensitive data access by cybersecurity researchers
• 78 percent of workers are self-teaching AI skills because employers fail to provide adequate training
• 46 percent of workers have spent over $1,000 of their own money on AI training

Youth AI Safety Institute says ChatGPT for Teens is unacceptable risk

Common Sense Media's Youth AI Safety Institute tested ChatGPT for Teens and rated it an unacceptable risk. The group ran over 4,000 prompts on accounts for 13- to 17-year-olds. It found that parents often did not get safety alerts during conversations about self-harm, suicide, or eating disorders. OpenAI disputes the findings and says the testing did not reflect how the product works.

Watchdog group says ChatGPT should be restricted for teens

A watchdog group says ChatGPT for Teens poses safety risks and should be limited. Common Sense Media tested the tool with more than 4,000 prompts on teen accounts. The group found that crisis responses were sometimes missing and that Study Mode could still help students cheat on homework. It wants better parental alerts, stronger crisis support, and blocks on homework completion.

OpenAI says teens can discuss suicide for hours before parents are alerted

OpenAI told Common Sense Media that parental alerts on ChatGPT for Teens can take several hours to activate for newly linked accounts. During tests, researchers found that safety notifications were rarely triggered during conversations about suicide or self-harm. OpenAI says linking teen and parent accounts securely can take time but alerts are sent within an hour once linked. Common Sense Media stands by its findings.

Nikesh Arora sees 40 companies racing to secure AI agents

Palo Alto Networks CEO Nikesh Arora says about 40 companies are being funded to secure AI agents, with at least 20 being Israeli-founded startups. Seventeen of those startups have raised about $900 million combined. He says the crowded market gives Palo Alto time to watch, learn, and later buy successful companies. AI agent security is a new and fast-growing area in cybersecurity.

2 cybersecurity stocks positioned for the AI agent boom

AI agents create new cybersecurity risks because they can access company systems and data with limited human oversight. Palo Alto Networks focuses on identity security through CyberArk, while CrowdStrike uses Falcon Guardian for runtime control. Gartner expects the market for securing AI to grow from about $2.8 billion in 2026 to $4.8 billion in 2027.

Elastic launches AlertZero AI agents for security operations

Elastic introduced AlertZero, a set of AI agents that automate parts of security operations center work. It handles alert triage, threat hunting, and rule adjustments. The product is in Technical Preview for Elastic Security customers. It works across Elastic Cloud, self-managed, and air-gapped deployments. Analysts keep control over which actions are automated.

MIMESIS trains user simulators for interactive AI agents

Researchers introduced MIMESIS, a user simulator trained on human conversations with 13 realistic behavioral patterns. The 9B model achieved a SOUL-Index of 65.7, outperforming frontier models. It improved behavioral fidelity by 13.4 points over Claude-Opus-5. Training agents with MIMESIS produced better results than using GPT-5.5 across eight environments. The study also proposed Coached On-Policy Self-Distillation for improved agent feedback.

Illusory pattern perception drives spurious inference in AI models

The article examines illusory pattern perception, a human tendency to find meaning in random data. This cognitive bias also appears in large language models. The study explores how this leads to spurious inference in AI systems. The research highlights parallels between human and machine pattern recognition errors.

Impel doubles UK dealer lead volumes with AI

Impel's Sales AI handled 519,982 UK automotive leads in the past year, doubling the previous year's total. The system has processed over one million leads across 300 UK dealerships. Customers include Vindis, Sytner Group, and Swansway. Sales AI generated 15,237 appointments and saved dealers 313,142 hours of time. The technology engages customers, books appointments, and hands off to dealership teams.

Applied Optoelectronics raises funds for AI networking growth

Applied Optoelectronics completed a $600 million equity offering of 5,694,845 shares at about $105.36 each. The company launched QuantumLink Central and QuantumWave 25G Tunable SFP for AI-driven network automation. The products target data center optics and cable infrastructure. Analysts project revenues of $5.5 billion and earnings of $1.2 billion by 2029. The main risk involves scaling new technologies while remaining unprofitable.

Reevo launches mobile AI sales app for on the go deals

Reevo released a new mobile app that helps sales people capture meetings and calls from their phones. The app records conversations, writes summaries, and updates deals without needing a separate meeting bot. It is available today on the App Store for iPhone and Google Play for Android.

AWS AI agent security flaws reveal autonomous agent dilemma

Cybersecurity researchers found that AWS repeatedly had to fix security holes in its autonomous AI agents. Agents were tricked into revealing credentials and could access sensitive data across the AWS region. The problem shows that the very access allowing agents to work also creates major security risks.

Most companies fail to provide AI training for workers

A TripleTen survey of 2000 workers found a big gap between what employers promise about AI training and what employees actually receive. While 47 percent of workers say AI changed their job, many have not been asked to learn new AI skills. As a result, 78 percent of workers are teaching themselves and 46 percent have spent over 1000 dollars of their own money on AI training.

Sources

• Keep Teens Off ChatGPT, Says Youth AI Safety Institute 
• Watchdog group says ChatGPT should be restricted for teens
• OpenAI Says Teens Can Talk About Suicide for Hours Before ChatGPT Will Notify Parents
• Nikesh Arora sees 40 companies racing to secure AI agents. At least 20 are Israeli
• 2 Cybersecurity Stocks Positioned for the AI Agent Boom
• Elastic launches AlertZero AI agent system for security operations
• MIMESIS: Learning User Simulators as Training Environments for Interactive Agents
• Illusory Pattern Perception Drives Spurious Inference in Large Language Models
• Impel doubles UK dealer lead volumes as AI use grows
• Applied Optoelectronics Stock Has AI Networking Demand Changed The Bull Case
• Reevo Launches Mobile AI Sales App to Keep Deals Moving Beyond the Desk
• AWS’s repeated problems with AI agent controls illustrates the autonomous agent dilemma
• Survey Results Reveal That Most Companies Are Failing to Provide AI Training