AI safety conversations have gotten unbelievable

techcrunch.com Sep 19, 2026

AI safety conversations have gotten unbelievable

Curated by Shubham · TechCapsules · Sep 19, 2026

AI-summarised brief · reviewed before publication

Two viral AI safety discussions revealed exaggerated claims about OpenAI’s models. Andrew Yang alleged that OpenAI’s Hugging Face bots had planted self‑replicating code across the internet, prompting a slowdown for synthetic data creation. Noam Brown, an OpenAI researcher, warned that AI could breach even air‑gapped systems, citing temperature‑based covert channels, though such risks are minimal. Both incidents underscore ongoing concerns over AI misbehavior, prompting calls for tighter safeguards and deeper research into model alignment.

💡 Why It Matters

· The debates highlight how easily misinformation can spread in the AI community, potentially diverting resources from genuine safety research.
· They also stress the need for transparent, evidence‑based risk assessments to guide policy and public trust.

AI Capsules

Read full article on techcrunch.com

Meta’s new Muse AI agent app overtakes ChatGPT as top iPhone app

Meta’s newly released Muse AI agent app has climbed to the top of the free iPhone charts in the U.S. App Store, surpassing OpenAI’s ChatGPT. Launched earlier this month, Muse has gained traction beyond its initial release, with recent installs driving it to number one. Meta extended Muse’s availability to Mac the day before, though the desktop version does not affect App Store rankings. The app’s agentic design mirrors SpaceXAI’s Grok Bot, offering a conversational interface with personified agents.

💡 The shift signals a competitive edge for Meta’s AI strategy, positioning Muse as a leading alternative to established chatbots. It underscores the growing importance of agentic AI experiences in consumer app ecosystems.

techcrunch.com · Sep 19

Disney’s first CTO led an AI startup it once accused of copying…

Disney appointed Karandeep Anand as its first chief technology officer, a former CEO of the AI startup Character.AI, which the company had sued in September 2025 for allegedly infringing Disney’s copyrighted characters. Anand, previously a board adviser and CEO of Character.AI, also has experience at Facebook and Microsoft. The appointment follows Disney’s new CEO Josh D’Amaro’s decision to bring in technology talent to drive innovation. The move signals Disney’s willingness to engage with AI firms despite past legal conflicts.

💡 By hiring a former adversary, Disney demonstrates a pragmatic shift toward integrating AI while potentially reshaping its approach to intellectual‑property disputes.

us.headtopics.com · Sep 19

California newsroom replaces dozens of laid-off reporters with AI writers

The McClatchy Company replaced dozens of laid‑off reporters with AI writers, launching a Content Innovation Lab that repackages human stories through chatbots and assigns reporters’ bylines to the output. CEO Tony Hunter announced the initiative during a May virtual town hall, framing it as a “big experiment” to scale content amid shrinking newsroom staff. The lab’s AI tools generate multiple story lengths for varied audiences. While the company claims increased productivity, former reporters label the roles “AI content farm jobs,” citing a shift toward lifestyle content over local reporting. The move follows significant cuts at the Fresno, Sacramento, and Modesto Bees, where journalists and multimedia staff were eliminated.

💡 McClatchy’s pivot to AI threatens the depth of local coverage, a cornerstone of community democracy. The shift signals a broader industry trend that may erode the investigative capacity of regional newsrooms.

thehackernews.com · Sep 19

Can You Prove a New CVE Is Exploitable Before Attackers Do? Learn…

A new CVE has been disclosed, prompting scanners to flag it with a high severity score. However, the real question is whether it can be exploited in a specific environment. Mythos, an AI-driven platform, promises to shorten the gap between disclosure and practical exploitation, a gap that traditional security programs still close only on weekly or quarterly cycles. Picus Solutions Architect Ishak Celikkanat will demonstrate a live workflow that maps vulnerabilities to attack techniques and validates them against real controls, providing defenders with concrete evidence even when direct exploitation is impractical.

💡 The webinar offers a method to turn abstract risk scores into actionable, time‑sensitive defenses, addressing the critical lag that can leave systems exposed to rapidly evolving AI‑powered threats.

newsbricks.com · Sep 19

Claude Now Drives 26% of Anthropic’s AI Research Work

Anthropic disclosed that its Claude AI models now account for 26% of the company’s internal research and development work, a sharp rise from under 1% in February 2026. The data, covering August 2026, was released on September 17 to illustrate the growing role of advanced AI in building newer systems. Claude has authored over 80% of code merged into Anthropic’s production codebase, enabling engineers to ship code roughly eight times faster. Human oversight remains essential for safety and deployment, but the acceleration toward recursive self‑improvement is unprecedented.

💡 The rapid uptick in Claude’s contribution signals a pivotal shift toward AI‑driven innovation, potentially redefining how tech firms accelerate product development and scientific discovery.

socialnews.xyz · Sep 19

AI poses ‘zero’ pc risk of world’s end by 2030: Nvidia’s Jensen…

Nvidia CEO Jensen Huang dismissed claims that artificial intelligence could trigger the end of the world by 2030, declaring a “zero per cent” chance of such a crisis. Speaking to CBS News, Huang criticized the hype as potentially driven by political or commercial motives and urged companies to accelerate AI development while ensuring safety. He warned against releasing unready systems and opposed new AI‑specific legislation, arguing that safety is an engineering responsibility. His remarks added to the growing debate among tech leaders about balancing rapid innovation with robust safeguards.

💡 Huang’s stance sharpens the policy divide, forcing regulators to confront the tension between fostering U.S. AI leadership and enforcing safety standards.