AI Safety Awareness Group Boston

​A free, virtual, public workshop — no technical background needed.
​What would it actually mean for machines to become smarter than humans — not just at chess or writing, but at everything? For decades, scientists and researchers have debated the idea of superintelligent AI: systems that could outperform humans across nearly all cognitive tasks. What once sounded like distant science fiction is now increasingly discussed as a real possibility within our lifetimes.
​This workshop is a guided, interactive introduction to superintelligence and the problem of loss of control — what happens if we create systems more capable than we are, but whose goals don't perfectly align with human values.
​Together, we'll explore:

• ​Superintelligence: What it means for an AI system to surpass human intelligence across domains, and how this differs from today's AI.
• ​Power and control: Why advanced AI could wield enormous economic, political, and strategic influence — and why "just turning it off" may not be realistic.
• ​Misalignment risks: How systems pursuing the wrong objectives could cause harm even without malicious intent.
• ​Societal choices: How decisions made today — by researchers, companies, and governments — may shape whether AI becomes a tool we control or a force we struggle to contain.

​What you'll get:

• ​A clear, non-technical introduction to core ideas from AI safety and existential risk research
• ​Live demonstrations of current AI capabilities to ground abstract concerns in real systems
• ​Small-group discussions and thought experiments connecting AI risks to everyday life, institutions, and incentives
• ​A better mental model for evaluating headlines, hype, and claims about AI progress

​Whether you're new to AI, casually interested, or deeply unsure what to believe, you'll leave with a clearer picture of why many researchers take superintelligence seriously — and what's actually at stake if we get it wrong.
***
​This is the final session in a three-part virtual series. If you missed Session 1 — Where is AI in 2026? (5/12) or Session 2 — The AGI Transition (5/14) — each workshop stands on its own and is free to attend.

I'm thrilled to continue this year's Rogue Agents workshops with a second technical, hands-on workshop in person at the Boston Public Library!

Content will be different from last time (where we studied Petri) and still geared toward practitioners in AI and Security who wish to learn more about analyzing, testing, and demonstrating agentic AI risks.

1. From 5.30-6.30, we'll highlight the OWASP Top 10 for Agentic Applications. To test for “rogue agent” behaviors in real time, we'll observe agents randomly changing their minds about which objects they impact and allowing their environment to hijack their goal through indirect prompt injections. We'll reserve a few minutes to discuss implications and real-world incidents.
2. From 6.30-7.30, we'll test the impacts of agents violating norms by spontaneously "solving problems", including captchas. We'll show how defenders can use this to defend web infrastructure, and how this tendency can cause worse security incidents.

To follow along step-by-step in Google Colab (less theory, more practice), please bring a laptop and set up an API key from an AI provider ahead of time (OpenRouter will work well). Some prior experience with python/and making LLM requests is recommended, but not strictly required if you’re comfortable learning by doing. All levels are welcome and encouraged to pair with someone at a different stage of learning.

The Library does not permit any outside food or beverage within the Study Rooms, and we abide by the Boston Public Library Code of Conduct. https://www.bpl.org/about-the-bpl/official-policies/appropriate-library-use/