Adobe doubles down on 'safe' AI

Adobe is pushing its artificial intelligence pitch away from novelty and toward legal cover, folding Firefly into video editing and marketing software while calling the tools safe for commercial use. (news.adobe.com) On February 12, 2025, Adobe introduced the Firefly Video Model and said it was “IP-friendly and commercially safe,” with Generate Video in the Firefly web app and Generative Extend in Adobe Premiere Pro. Adobe said Firefly had already been used to generate more than 18 billion assets globally. (news.adobe.com) On April 2, 2025, Adobe moved that video model deeper into Premiere Pro by making Generative Extend generally available, with support for four-kilobyte video and vertical video, plus Media Intelligence to search large footage libraries and Caption Translation in 27 languages. (news.adobe.com) Adobe widened the same message on March 18, 2025, at Adobe Summit, where it launched Adobe Experience Platform Agent Orchestrator for businesses to build and manage artificial intelligence agents across Adobe and third-party systems. Adobe said the system would support customer experience and marketing workflows with a suite of 10 purpose-built agents. (news.adobe.com) That matters because Adobe is selling to companies that worry less about one-off image generation than about copyright claims, privacy rules and whether a tool fits into software they already use. Adobe’s March 18, 2025 Summit announcement tied those agents to first-party data, content workflows and customer journey tools inside its enterprise stack. (news.adobe.com) In Europe, Adobe has paired that product push with a compliance message. On September 25, 2024, Adobe said it had signed the European Union Artificial Intelligence Pact and backed guardrails around transparency and tools to combat deceptive synthetic media. (blog.adobe.com) Adobe has repeated that safety framing for more than a year. In March 2024, the company said it had built Firefly around “commercial safety,” creator protection and policy work on deepfakes, extending a line it had also used in 2023 when it backed White House voluntary commitments on artificial intelligence. (blog.adobe.com, blog.adobe.com) Adobe’s sales pitch also became more complicated in 2025. At MAX London on April 24, 2025, Adobe said Firefly would include not only Adobe’s own models but also partner models from Google Cloud and OpenAI, even as it continued to describe Adobe’s Firefly models as commercially safe. (news.adobe.com) Critics have argued that this creates a sharper distinction between Adobe’s in-house models and outside models offered through the same interface. Creative Bloq wrote in 2025 that Adobe’s long-running “commercially safe” promise sat uneasily beside the addition of partner systems trained under different rules. (creativebloq.com) Adobe’s answer has been to make safety a product feature, not just a policy claim: video generation inside Premiere Pro, agent software inside marketing tools, and a steady emphasis on licensed training data, enterprise workflows and compliance language. (blog.adobe.com, news.adobe.com)