As AI models have become more powerful, the potential for those models to be misused has grown \u2014 as has a clamor for safety guardrails that can stop such abuse from happening. AI companies must now walk a delicate tight rope between respecting their enterprise customers\u2019 privacy while also watching usage for possible issues.<\/p>\n
Sensing an opportunity to one-up its rival Anthropic, OpenAI just announced a privacy-centric safety approach to monitoring for misuse. The company is previewing<\/a> a new service to select customers that it calls Private Safety Processing. This is an automated system that watches for potential abuse while simultaneously retaining none of the customer\u2019s data.<\/p>\n This system clearly runs counter to Anthropic\u2019s recently announced<\/a> data retention policy. The policy, which has aggravated some customers, enables the AI lab to keep user data (all of their sessions \u2014 and the conversations therein) for a period of 30 days, when it comes to \u201ccovered models.\u201d Those models include all Mythos-class models and \u201cfuture models with similar capabilities,\u201d the company says.<\/p>\n This policy, which was announced in July, was designed for the purposes of safety allowing the lab to sift and analyze potential impropriety. However, it has deeply concerned<\/a> some enterprises that handle large amounts of sensitive data and don\u2019t want it harbored (or inspected) by the AI lab.<\/p>\n OpenAI \u2014 like most other AI companies \u2014 already afford customers a relative level of privacy by adhering to a policy known as Zero Data Retention. ZDR uses agents within the OpenAI API to monitor for abuse on a per session basis. In this way, customer data isn\u2019t retained by the company but companies are still able to scan for bad activity without the need for human intervention. It\u2019s worth noting that Anthropic also largely abides by ZDR \u2014 except when it comes to \u201ccovered models,\u201d like Fable.<\/p>\n OpenAI says that Private Safety Processing is a new technology that widens ZDR\u2019s scope. It describes it as a form of long-horizon safety monitoring that assesses the inputs and outputs of multiple conversations \u2014 not just one. Again, the monitoring is conducted by an agent, which, if triggered, catches interactions and analyzes them across sessions for signs of potential misuse. <\/p>\n The new tech helps OpenAI detect malicious use of AI that takes place over multiple sessions, a spokesperson told TechCrunch. A bad actor \u2014 hypothetically someone trying to engineer malware for a cyberattack \u2014 may spread out their requests to avoid detection. Private Safety Processing can analyze those multiple conversations for signs of abuse without human review of a user\u2019s conversations.<\/p>\n In the case where the system is triggered, it may send a \u201cnarrowly defined signal\u201d to OpenAI that warns of a specific type of activity, the company says. Based on that signal, OpenAI can then decide whether \u201cenforcement is necessary,\u201d it says. If so, OpenAI will reach out to the customer for more context or to work with them on the issue and a customer may choose to share data with OpenAI at their discretion, the spokesperson said. <\/p>\n By contrast, Anthropic notes that human review of customer data can occur, but only \u201cthrough a controlled access path\u201d that involves \u201ca small set of approved reviewers.\u201d Every one of those review sessions is \u201crecorded in a tamper-proof log that reviewers cannot suppress or modify,\u201d the company says. <\/p>\n The corporate competition between OpenAI and Anthropic is tense at the moment, with both companies looking for any opportunity to gain an advantage on the other. A recent report<\/a> showed that OpenAI\u2019s Q2 grew more slowly than Anthropic. Anthropic\u2019s annualized revenue run rate is now reportedly $65 billion<\/a>. Anthropic investors have said it could IPO at $2 trillion<\/a>, while OpenAI is also working on its IPO.<\/a><\/p>\n