AAAI 2027 workshop on trustworthy AI agent evolution lands Bengio, Wang, Du as speakers
berkeley_ai · x · 2026-10-17
The Trust4RSI-Agent workshop at AAAI 2027 (Feb 22–23, 2027, Montreal) tackles trustworthy evolution of AI agents through an agent operating system lens.
- Core thesis: agents are shifting from static pretrained systems to adaptive ones that update models, memories, tools and architectures; in long-horizon workflows small errors accumulate and amplify. Many trust failures are systems failures — memory corruption without provenance, alignment drift from irreversible updates, unsafe tool calls lacking permissions and sandboxing.
- Topics: memory evolution (provenance/isolation/rollback), tool & skill evolution (permissions, sandboxing, verification), model & policy evolution (versioning, validation), multi-agent co-evolution, and evaluating trustworthy evolution (tracing, monitoring, benchmarking).
- Speakers/panelists: Yoshua Bengio, Mengdi Wang, Yilun Du, Hugo Larochelle, Bill Zhu, Yue Zhao.
- Timeline: papers due Nov 20, 2026; notifications Dec 2, 2026; full papers 8 pages, short/position 4 pages, double-blind via OpenReview.
More from coding & agent
- Latent Space's High Taste Testers: power users compare Dots, Muse, GrokBot, Instinct — altryne · 2026-10-17
- Agent blocked by auth: asking users to paste email sign-in codes sparks workflow complaints — carlosdponx · 2026-10-17
- Opus 5.5 built a working computer: 277k logic gates in JavaScript with an OS and a game — FinanceYF5 · 2026-10-17
- Grep Launches Upgraded MCP to Build and Run Agents from Claude, ChatGPT and Codex — garrytan · 2026-10-17
- Cursor's mobile app lets you babysit your AI agents from anywhere, even weekends — nateliason · 2026-10-17
- NoHumans.Space seeks testers for a knowledge network where AI agents share findings — dark-epiphany · 2026-10-17