Dimension scores are derived from public data and fields; weighted into the composite. Reference only.
ControlConf 2026 is a two-day in-person conference focused on AI control, held in Berkeley, California, on April 18–19, 2026. Here, AI control is defined as reducing the risk of AI misalignment through interventions that remain robust even when a model attempts to undermine safeguards. The event also includes a pre-conference Control Workshop on April 17, aimed at participants who are relatively new to AI control and want to quickly build the necessary background knowledge.
From the perspective of a course or educational product, this is not a standard recorded course or live bootcamp, but rather a research conference with an accompanying workshop. The main conference includes talks, fireside chats, and breakout discussions, with the goal of sharing research progress, raising open problems, and encouraging collaboration between labs and research organizations. The workshop includes talks, reading groups, and tabletop exercises, making it closer to an intensive introductory training session. The organizers are Redwood Research and FAR.AI, with publicly listed guests from institutions such as Redwood Research, METR, Anthropic, and CMU, giving the event a strong academic and industry profile. The publicly available materials do not specify the language of instruction, nor do they mention any accreditation or certificate.
Both the conference and the workshop are free to attend, which is a clear advantage. However, registration is not an open ticket purchase: applicants must submit an application, which the organizers review, with feedback expected within two weeks. Applicants invited to the workshop will automatically receive admission to the main conference. As a result, the real barrier is not cost, but research background, fit with the event’s focus, and the application screening process.
The main advantage is that the topic is highly cutting-edge, making it suitable for people who want to enter AI control, AI safety, and AI risk modeling and gain access to first-hand discussions. It also offers opportunities to connect with researchers, engineers, and policy professionals across organizations. The downsides are that the learning structure is not like a systematic course, and the public materials do not indicate any post-event mentoring, assignment feedback, or certificate. Since the event is in-person in Berkeley, participants from China also need to consider visas, scheduling, and travel costs.
It is best suited to researchers and practitioners with backgrounds in AI safety, machine learning, security, formal methods, or policy, as well as junior researchers hoping to enter AI control through the workshop. Accessibility from China cannot be determined from the available text; basic access to the official website is unknown, while actual attendance would mainly depend on international travel and the application process. If attending is not feasible, following FAR.AI, Redwood Research, and related AI safety conferences or public research talks may be useful alternatives.
⚠ This review is compiled from public sources and does not constitute a purchase recommendation. Verify all facts on the vendor's official site. Verify on controlconf.org official site.
controlconf.org is an United States Education provider. TG4G tracks its product information, an overall rating of 7.0/10, and a China-accessibility score of Workable. Click "Visit Official Site" to reach controlconf.org directly.