Security

Training AIs to Solve CAPTCHAs

Read full article at source ↗

Based on a feed excerpt; full article content has not been acquired. Source evidence retrieved 19/10/2026, 11:15:24 UTC.

Briefings

Research shows vision-language models (VLMs) can solve visual CAPTCHAs without prior task-specific training, yet each new challenge requires full retraining. The study pairs a fine-tuned YOLOv8 detector with an open-weight VLM using only screenshots and OS input events. The system achieves 85.4% overall and 84.2% macro accuracy across 16 classes, recovering from zero accuracy after perturbation attacks. A cheap open-weight teacher can harden the solver as effectively as a perfect oracle within a simulated arms race.

Key points

• Vision-language models solve visual CAPTCHAs without task-specific training, but each new challenge requires full retraining from scratch.
• The system achieves 85.4% overall and 84.2% macro accuracy across 16 classes using only screenshots and operating-system input events.

Context

This research investigates how adaptive solvers improve with use, contrasting with static detectors that invert the trade-off between speed and accuracy. The findings are critical for understanding the resilience of CAPTCHA defenses against evolving bot strategies.

Limitations

The study relies on simulated arms race data and lacks long-term field validation. Open-weight teacher accuracy is not explicitly measured, and the specific perturbations used in the simulation are not detailed.

Generated by Bittrees-hosted qwen35-2b sha256:aaf42c8b7c3c on 19/10/2026, 11:31:26 UTC. Based on publisher-provided feed evidence; may not cover the full original article. Check the source for details.

View archived briefing on IPFS
bafybeic4gwu4zjxjp7fhf62xi4e6gcrzxqiqudbtekqrelnrtellli2ayy