Anthropic Cuts Rival AI Data Harvesting by 90% to 99%

The safeguards have turned away trillions of tokens of paying traffic and meaningful revenue, though the company began charging for certain blocked extraction requests on September 24. While the current classifier has a false-positive rate below 0.1%, some users report that legitimate evaluation runs have been flagged. Anthropic noted its defenses have not yet faced a highly sophisticated or persistent attempt to bypass them, but the results suggest technical defenses may be more effective than the government regulations previously sought to protect intellectual property.