What’s new
This quarter, the most interesting AI shifts are about changing human behaviour. AI is transitioning from simple content generation to autonomous action, shifting from open experimentation toward more controlled access, and evolving from a private tool into a fundamental part of the media and search landscape.
The following AI developments also caught our attention:
AI doing
The big shift this quarter is that AI is becoming less like a clever chat box and more like a junior colleague with tools. OpenAI’s GPT-5.5 push, Claude Cowork, and Claude Code all point in the same direction: agents that can work across files, code, apps and systems, not just generate a neat paragraph. That is useful, but it also changes the risk profile. Once AI can click, edit, run, publish and send, governance can’t just be a PDF of principles. It needs permissions, review points and proper accountability. Less “what can it say?” and more “what can it actually do?”
Claude Mythos and the governance problem
Claude Mythos is a new LLM model that Anthropic says is its most capable yet for coding and agentic tasks, with particular strength in cybersecurity. It has been given restricted access rather than a broad public release. Why? Because the same capability that helps you find cybersecurity vulnerabilities can also help you exploit them. Anthropic says it has already helped identify more than 10,000 high or critical severity vulnerabilities in systemically important software. That is impressive, but also uncomfortable. In the right hands, this helps organisations find and fix serious weaknesses faster. In the wrong hands, it could make it easier to discover and attack those same weaknesses. This is not just about making a powerful model available; it is about controlling who can use it, and what they can use it for.
ChatGPT Ads go live
What once felt like a distant plan is now starting to happen. OpenAI has begun testing paid contextual ads in ChatGPT for Free and Go users, clearly labelled as “Sponsored” and separated from the assistant’s response. The interesting bit is the targeting: not just keywords or demographics, but context hints that match advertisers to relevant conversations. For brands, the shift is clear. They can earn search, media and brand visibility by being relevant and credible enough to appear in answers, or buy sponsored space around those conversations.
Our approach to using Generative AI – Q2 2026
As regulation and creative possibilities continue to evolve rapidly, this approach reflects where we are today, shaped by experimentation, close collaboration with clients, and ongoing industry developments.
Generative AI is no longer speculative. It is embedded across ideation, prototyping, and production, and increasingly within live, customer-facing experiences. We’re sharing our approach openly to encourage collective learning, dialogue, and co-creation across the creative industry, while remaining mindful of the legal and ethical considerations involved.
Recognising that every project and client is different, this approach tries to provide guidance and clear boundaries without being prescriptive, leaving room for judgement, craft, and creativity.
We will treat Gen AI as a tool to extend our creativity
It’s not a replacement for our creativity, but it can allow us to work more efficiently, explore more broadly, communicate more richly and create more value for our clients. Wherever appropriate, we will be unafraid to use it to enhance our work.
AI will not be a substitute for our original work
We will never rely solely on AI-generated solutions; we will ensure that the final deliverable is the product of our creativity. We will ensure that our AI outputs are original and distinct, avoiding inputs that mimic or replicate third-party content or aim to create in another’s style. This maintains the IP, authenticity, uniqueness and value in our work.
We will not use AI as a source of truth
Generative AI is known to lie or “hallucinate”, stating fiction as though it is fact. We will always evaluate the outputs of generative AI with a critical eye and not rely on it as a single source of truth.
We will be transparent about how and when we use AI in our work
We will be open about how we use AI in our work. We will not use AI outputs as final, client- and market-facing deliverables without both written client and Imagination Legal team permission. As with all our work, wherever possible, we will give proper credit and obtain permission for any artist’s, musician’s or author’s work that we incorporate.
We will limit the information we share on open platforms
We know that many AI tools are cloud-based and may reuse the data we provide. We will not share personal data (including a person’s image or voice), project-specific data such as images or identifiable text from past projects, confidential information, or third-party intellectual property without consulting with the Imagination Legal team.
We will experiment, share and learn
We will share different tools, ideas, prototypes, successes and failures so we can all learn and improve, including which tools, prompts, and methods we are using to get the most out of these tools. We will ensure that anyone who is using AI tools for client work will be given the appropriate guidance and training.
We will be proactive in recognising and mitigating bias in AI
We create fair and inclusive work. We will be attentive to potential biases in AI outputs such as ethnicity, age or gender, and we will engineer prompts or other inputs to help mitigate this. Good input in, good input out.
We apply additional care when using generative video, voice, and AI agents
Generative video, synthetic voice, and conversational agents introduce heightened ethical, legal, and reputational considerations. When using these tools, we will be particularly mindful of consent, likeness and voice rights, transparency, and brand trust. We will not create or deploy synthetic representations of real people’s faces, voices, or identities without explicit permission and appropriate legal approval. Where generative video, voice, or agents are used in client-facing or public contexts, we will ensure their use is intentional, disclosed where appropriate, and aligned with the values of both Imagination and our clients.
We know the landscape of this emerging technology is constantly shifting, and we’ll be posting more of our latest thinking, proof of concepts and insights. If you have any questions on the above or want to talk to us about running Gen AI workshops for your teams, please contact us at info@imagination.com.
More like this
