Anthropic announced on Tuesday (1) a change to its controversial data retention policy for corporate clients, in response to various user feedback. As a replacement, the company plans to introduce a new system called Enterprise Frontier Safeguards.
The previous retention policy was introduced in June, coinciding with the launch of the company's most sophisticated artificial intelligence (AI) models, Claude Fable 5 and Mythos 5. At that time, Anthropic stipulated that all traffic from these models would need to be retained for a period of 30 days, aiming to assist in combating cyberattacks and 'complex and new' misuse.
Although the company guaranteed that the data would not be used for any purpose other than security, including model training, many enterprise clients expressed apprehension about this determination.
Kate Jensen, Anthropic's head for the Americas, reported that the organization invested hundreds of hours collaborating with clients to develop a viable alternative.
Founded in 2021, Anthropic bases a large part of its revenue on selling services and products to businesses. The corporate sector is seen as crucial for the company as it prepares for a potential Initial Public Offering (IPO).
Anthropic also competes in the market with other companies, such as OpenAI, which launched its own zero-data retention solution for advanced models earlier this month.
Last week, Dario Amodei, Anthropic's CEO, mentioned Enterprise Frontier Safeguards during an interview with CNBC, addressing the expansion of the company's partnership with Salesforce. Amodei explained that this solution ensures the privacy of Salesforce customers' data and prevents Anthropic's models from 'getting out of control.' He added that the company dedicated an 'enormous amount of effort' to managing the permissions of the integration between the two companies.
Jensen expressed great confidence that the announcement will result in an acceleration of the adoption of Anthropic's products, particularly in activities involving more sensitive customer information.
Additionally, the company released its two newly launched models on Tuesday, Claude Fable 5.1 and Mythos 5.1, which, according to Anthropic itself, show notable improvements in areas such as programming, knowledge, and research.
Enterprise Frontier Safeguards will be implemented gradually, and Anthropic plans to expand its availability throughout the autumn in the Northern Hemisphere.