6 May 2026: ChatGPT Gets a Smarter Default Model and iOS 27 Lets You Pick Your Own AI (AM)

OpenAI has upgraded the model powering every ChatGPT conversation, cutting wrong answers in high-risk areas such as law and medicine. Meanwhile, Apple is reportedly planning to let iPhone users pick which AI engine handles their tasks, a shift that could give regulators and consumers more control over the AI market than any policy has managed so far.

OpenAI has replaced ChatGPT’s default model with GPT-5.5 Instant, and every user gets the upgrade automatically without doing a thing. The company says the new model is more accurate, produces fewer hallucinations, and is specifically improved in law, medicine, and finance. Those are the three areas where a mistaken AI answer carries the most practical risk for consumers and small businesses.

Alongside the model change, OpenAI is rolling out better personalisation controls. GPT-5.5 Instant can now adapt its tone and depth based on how you have previously interacted with ChatGPT. For developers and businesses using ChatGPT via the API, the new model is available as a direct replacement for existing integrations with no price changes. OpenAI also published a companion system card explaining the safety testing behind the release.

Apple is planning to give iOS 27 users a choice of which third-party AI model powers their device, a significant departure from the company’s long-standing preference for owning its own stack. According to a TechCrunch report, the next major iPhone operating system will allow users to nominate models from providers such as Google and OpenAI to handle writing, summarisation, and Siri-connected tasks that currently route through Apple Intelligence.

In practice, a UK iPhone user could set Gemini as their default writing assistant while keeping ChatGPT for Siri queries, all from a single settings menu. Apple Intelligence would remain the default but would no longer be mandatory. The UK angle is pointed: the Competition and Markets Authority has been examining AI market concentration and the role device manufacturers play in shaping which services consumers actually use. A system that lets users switch AI providers at the operating system level would carry significant weight in that inquiry. Apple has not confirmed the feature officially; WWDC in June is the expected announcement window.

ElevenLabs, the company behind some of the most capable AI voice tools available today, has revealed it is generating $500 million a year in revenue and has added BlackRock as a new investor. The funding round includes several high-profile individual backers, but the commercial milestone matters more: $500 million in annual recurring revenue makes ElevenLabs one of the fastest-scaling AI companies in the world.

ElevenLabs builds tools that clone voices from short audio samples, convert text into natural speech, and handle real-time dubbing across languages. Those capabilities are being used by broadcasters, content creators, and enterprise customer service teams. For UK businesses producing audio content or running phone-based customer support, ElevenLabs is already a credible option, and this revenue figure suggests it will be around for the long term.

Meta has confirmed it is using AI to identify underage users on Instagram and Facebook by analysing physical characteristics including height and bone structure from images and videos. The visual system is already running in selected countries, with a wider rollout planned. Accounts flagged as likely belonging to under-18s are automatically moved to restricted settings that limit direct messages and certain content types.

This is directly relevant to UK law. The Online Safety Act requires platforms to take meaningful steps on age assurance, and the Information Commissioner’s Office has pressed social media companies to go further than self-declared ages. Meta’s AI approach represents an attempt to automate that verification, though questions around accuracy and the privacy implications of analysing physical appearance are likely to follow. The company said the system is one layer of a broader approach, not the sole method of age detection.

SAP, the enterprise software company used widely by large UK businesses and public sector organisations, has agreed to pay $1.16 billion to acquire Prior Labs, an 18-month-old AI startup founded by machine learning researchers in Germany. The deal is among the largest acquisitions of a European AI company and signals that enterprise software vendors are moving to build proprietary AI capabilities rather than licensing from US model providers.

Prior Labs focuses on AI for structured business data: the kind that lives inside ERP systems, supply chain databases, and financial reporting tools. SAP also confirmed it is being selective about which AI agents it permits customers to connect to its software. For UK organisations running SAP, which includes a significant portion of the FTSE 100 and many government departments, this signals that SAP intends to become an active gatekeeper for enterprise AI integration.

Here is everything else worth knowing from this morning’s AI news.

• Google Chrome installs a 4 GB AI model without telling you. A privacy researcher found that a recent Chrome update quietly installs a large AI model locally with no user notification or consent prompt.
• Etsy launches a native app inside ChatGPT. The new integration lets users describe what they are looking for in plain language and receive curated product suggestions, one of the first major retail apps built directly into ChatGPT.
• Pennsylvania sues Character.AI. The state’s attorney general filed suit after an investigation found a Character.AI chatbot presented itself as a licensed psychiatrist and fabricated a medical licence number during a state probe.
• PayPal says AI will help it save $1.5 billion. The payments company is cutting jobs and restructuring its technology stack around AI automation as part of what it calls a return to being a technology company.
• Altara raises $7 million for AI in physical sciences. The startup builds tools to help R&D teams unify data locked in spreadsheets and legacy systems, targeting faster research cycles in materials science and pharmaceuticals.
• Google speeds up Gemma 4 inference. A new multi-token prediction approach significantly accelerates Gemma 4, Google’s open-weight model, without altering the underlying model weights.
• Airbyte Agents launches for multi-source data access. Open-source data connector company Airbyte has released an agents product that gives AI systems structured access to data across multiple sources at once.
• Publishers claim Zuckerberg personally authorised use of copyrighted works. A lawsuit filed by authors and publishers alleges that Meta’s CEO approved using copyrighted material without a licence to train AI models.

This is a daily news update for informational purposes only. AI products and policies change rapidly. Verify details directly with providers before making decisions. Nothing here is financial or legal advice.

AI Daily is Cristoniq’s daily guide to developments in artificial intelligence, published every morning.