OpenAI launches GPT-Live

OpenAI has introduced GPT-Live, a new generation of voice models that it says brings ChatGPT closer to natural human conversation than any previous release. The system replaces the turn-based approach of Advanced Voice Mode with a full-duplex architecture, allowing the model to listen and speak simultaneously, interject with acknowledgements, and decide multiple times a second whether to talk, pause or stay quiet.

Perhaps the more significant change is architectural rather than conversational. GPT-Live handles the flow of dialogue itself, but hands off anything requiring web search, complex reasoning or agentic work to GPT-5.5 running behind the scenes, then folds the result back into the conversation without breaking the exchange. OpenAI frames this as decoupling interaction from intelligence, meaning the voice layer can keep pace with future frontier models without being rebuilt each time.

Two versions are rolling out globally on iOS, Android and the ChatGPT web app: GPT-Live-1, becoming the default for paying tiers, and a mini variant for free users. Reasoning effort can be dialled between Instant, Medium and High. OpenAI reports gains over Advanced Voice Mode on GPQA (expert-level scientific reasoning), BrowseComp (agentic web search) and an internal telecom-support benchmark, τ³-Voice Telecom.

Safety testing was expanded to cover self-harm, psychosis and mania, emotional reliance on AI, violence and sexual content, with built-in safeguards able to steer, message or end conversations judged higher risk, alongside dedicated protections and parental controls for teen users. Video and screen sharing are not yet supported.

AI for Leaders: Governance, Risk & Why Strategy Must Come First

Is your organisation adopting AI without a defined strategy or governance? If so, you’re not alone — and the consequences range from data breaches to reputational damage to safeguarding failures. AI is a strategic risk, not a silver bullet, and definitely not a magic data fairy. Without clear ownership, rules, and accountability, AI can proliferate across your organisation in ways that cost more — in time, money, and trust — than not using it at all. Speakers: Craig Clark, Director at Clark & Company Information Governance Services, and Stewart Tinson, Project Director, AI-360. You’ll learn: Why “understand your why” must precede any AI rollout Which risks senior leaders most commonly overlook — from data leakage to safeguarding How to measure AI value through outcomes, not adoption metrics Why accountability for AI governance sits with the C-suite, not IT or compliance Key topics: AI strategy • Risk appetite • Data protection • Bias & fairness • Safeguarding • AI literacy • Governance models • Return on investment • Executive accountability • AI use case tiering For CISOs, CIOs, and senior leaders who need to govern AI responsibly — before it governs them.

BrightTALK