21 April 2026: OpenAI Launches GPT-Image-2, Adobe Goes Agentic, FCA Names Eight Banks for AI Testing (AM)

OpenAI unveiled its next-generation image model on Tuesday morning, Adobe turned its annual Summit into an agentic AI showcase, and the FCA confirmed eight major financial firms are now live-testing AI under regulatory supervision. Three announcements from the same day, each pointing in the same direction: AI is moving from prototype to production.

OpenAI launched GPT-Image-2 today, replacing DALL-E with a model that renders text inside images at near-perfect accuracy and generates at twice the speed of its predecessor. The announcement came alongside a noon PT livestream on 21 April, with the company describing the new model as its most capable image synthesis system to date. The headline improvement is typography: GPT-Image-2 achieves a 99 per cent accuracy rate when placing readable text inside generated images, an area where every major AI image tool has historically struggled. A poster that says what you want it to say, a mock-up UI with correct labels, a logo treatment that holds up at any size. These are tasks that previously required a human designer to finish what the AI started.

The model outputs images at 4,096 by 4,096 pixels and does so roughly twice as fast as GPT-Image-1. OpenAI also embedded next-generation watermarking and content provenance classifiers directly into the output, which matters for publishers, advertisers, and anyone using generated images in public-facing material. The rollout starts today for ChatGPT Plus, Team, and Enterprise subscribers, with API access scheduled for early May. For consumers and small businesses that rely on ChatGPT for creative work, this is a meaningful practical upgrade available right now.

Adobe used the second day of its annual Summit on 21 April to launch CX Enterprise, an end-to-end agentic AI system for customer experience management, built in partnership with NVIDIA and WPP. The product centres on CX Enterprise Coworker, which Adobe describes as an agentic layer that can plan, execute, and adapt customer experience workflows across the full customer lifecycle, from acquisition through to retention. A live keynote demo ran on NVIDIA’s Agent Toolkit, using its OpenShell secure runtime and Nemotron open models as the underlying infrastructure.

The practical pitch is that marketers at enterprise scale can now orchestrate campaigns, personalisation, and customer data pipelines without writing code for each handoff between tools. Major agency groups including Omnicom, Publicis, Havas, and WPP have committed to standardising on CX Enterprise for joint client work. For small businesses, the immediate relevance is indirect: if the agency your competitors use adopts this, the gap between what large and small marketing teams can produce will widen. Adobe has not confirmed pricing or small-business access tiers for CX Enterprise Coworker.

The FCA confirmed on 21 April that eight firms have joined its second AI Live Testing cohort, with the announcement timed to UK FinTech Week. The participants include Barclays, Experian, Lloyds Banking Group through its Scottish Widows subsidiary, and UBS, alongside GoCardless and three others. The cohort spans a broad range of AI applications, from agentic AI systems and small language models to experimental neural architectures. Jessica Rusu, the FCA’s chief data, information and intelligence officer, said the selection reflects the fast-evolving nature of AI models in financial services and the need for direct regulatory involvement in testing before deployment.

The FCA also confirmed it will publish a good and poor practice report on AI in financial services later in 2026, which will set out what regulators expect to see and what they do not. For anyone building or using AI tools in UK financial services, that report will carry real weight. The Live Testing programme gives the FCA direct visibility into how these models behave in production conditions, rather than relying solely on what firms self-report. This is the most substantive AI regulatory action in the UK financial sector since the first cohort launched, and it signals that the FCA is moving from observation to active governance.

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

• Amazon commits up to $25 billion more to Anthropic. A follow-on to its earlier investment, with Anthropic committing to spend more than $100 billion on AWS infrastructure in return. CNBC [20 Apr]
• Claude Opus 4.7 launched. Anthropic’s latest model brings a 13 per cent coding benchmark improvement, enhanced vision, and new cybersecurity safeguards. Anthropic [16 Apr]
• Claude Design by Anthropic Labs. A consumer-facing design tool powered by Opus 4.7, available to Pro, Max, Team, and Enterprise users at claude.ai/design. Anthropic [17 Apr]
• Google DeepMind Gemini Robotics-ER 1.6 released to API. A robotics-specific model update now accessible to developers through the DeepMind API. [15 Apr]
• NVIDIA Ising open AI models for quantum computing. A 35-billion-parameter vision-language model that automates quantum processor calibration, reducing a days-long process to hours. NVIDIA [14 Apr]

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.