Is it just \u201cfear-based marketing\u201d?<\/p>\n
The new results for GPT-5.5 suggest that, when it comes to cybersecurity risk, Mythos Preview was likely not \u201ca breakthrough specific to one model\u201d but rather \u201ca byproduct of more general improvements in long-horizon autonomy, reasoning, and coding,\u201d AISI writes.<\/p>\n
In a recent interview with the Core Memory podcast<\/a>, OpenAI CEO Sam Altman criticized what he calls \u201cfear-based marketing\u201d in promoting limited releases for certain AI models. While he said he\u2019s \u201csure Mythos is a great model for cybersecurity,\u201d he added that \u201cit is clearly incredible marketing to say, \u2018We have built a bomb. We are about to drop it on your head. We will sell you a bomb shelter for $100 million.\u2019\u201d<\/p>\n \u201cThere will be a lot more rhetoric about models that are too dangerous to release,\u201d Altman continued. \u201cThere will also be very dangerous models that will have to be released in different ways.\u201d<\/p>\n In February, OpenAI rolled out its Trusted Access for Cyber pilot program<\/a>, letting security researchers and enterprises verify their identities<\/a> and register their interest in studying OpenAI\u2019s frontier models for \u201clegitimate defensive work.\u201d Last month, OpenAI said it was using that trusted access list to control the limited launch of GPT-5.4-Cyber<\/a>, a model variant that it says is \u201cpurposely fine-tuned for additional cyber capabilities and with fewer capability restrictions.\u201d<\/p>\n