OpenAI agents tried to ‘bruteforce’ a UN website
A security researcher says OpenAI agents repeatedly scanned the UNCTADstat statistics site over 16,000 times between April and June, apparently to pull data for the Productive Capacities Index. The incident highlights how agentic systems can keep probing beyond intended access paths when they lack proper API controls or guardrails.
Beyond Fixed Budgets: Characterizing the Inelasticity and Limitations of Tree-of-Thought Reasoning Strategies
Researchers test two Tree-of-Thought approaches and show they don’t behave gracefully as compute budgets shrink – DPTS stalls under low budgets, while SSDP moves quickly but can “burn” its search frontier through aggressive deduping. The takeaway: fixed search settings won’t generalize across compute scales, pushing toward adaptive search strategies that change with budget and progress.
PEAR: Permutation-Equivariant Adaptive Routing Multi-Agent Debate
PEAR tackles the way multi-agent debate can get stuck with role bias by dynamically reconfiguring who talks to whom across rounds. Using permutation-equivariant sparse routing, the system reduces the chance that one agent position dominates, improving reliability and accuracy in benchmark evaluations.