GPT-5.3-Codex: Сэм Альтман выдал хакерам «черную метку»

OpenAI has finally stopped pretending that its language models were created solely for writing poetry and summarizing long letters. Sam Altman has officially unveiled GPT-5.3-Codex, and this is far more than just another minor update.

This is a full-fledged claim to dominance in the cybersecurity sphere. In an industry where a single code error can cost billions, OpenAI has decided to take on the role of chief security architect and defender of digital borders. The real news here isn't even the model's release itself, but its certification.

GPT-5.3-Codex became the first model from the company to receive a "high-level" rating under the network measurement protection system. For a long time, critics have pointed out that AI assistants often hallucinate vulnerabilities or, worse, help aspiring hackers write malware.

Now OpenAI claims to have created a true digital bunker, capable not only of writing code but of spotting its weak points before malicious actors can exploit them. To cement their success, the company is implementing a pilot program called "Trusted Access Systems." This sounds like the beginning of a bureaucratic nightmare, but in reality it means the end of the "Wild West" era in the use of powerful algorithms.

OpenAI is no longer going to hand out its sharpest tools to just anyone without discrimination. To gain access to Codex's most sensitive functions, developers will need to prove their trustworthiness and pass through the filters of the new system. This is a logical, though somewhat painful, step toward a closed but secure ecosystem.

The financial aspect of this launch deserves separate attention. Altman didn't just release a product—he backed it with a $10 million check in the form of API credits. This money will go toward supporting those building defensive systems and seeking ways to strengthen digital infrastructure.

OpenAI is effectively subsidizing "white hat hackers" and security specialists, attempting to create an army of loyal allies around itself. This is a smart strategic move: if the next global attack is prevented using their technologies, the reputational dividends will outweigh any costs from grants. Of course, such claims should be viewed with a healthy dose of skepticism.

In the world of information security, the status of an "impregnable fortress" typically lasts only until the first major hacker conference. By calling its model "high-level" in terms of protection, OpenAI has essentially painted a giant target on it. The research community will view this as a personal challenge, and the real stress test of GPT-5.

3-Codex will begin not in the company's sterile labs, but in the first hours after access opens to the pilot group. The release of this model also signals an important shift in the company's strategy. The era of universal models that can do "a little bit of everything" is gradually giving way to the age of narrow specialists.

GPT-5.3-Codex is not designed to create workout plans for you or suggest dinner recipes. Its job is to ensure your database doesn't end up on a dark forum the next morning.

This is a more mature, pragmatic, and apparently more profitable path forward for a company that wants to become indispensable to major corporations. Ultimately, we're witnessing how AI giants are beginning to battle for critical infrastructure. If previously Twitter debates were about whose model tells better jokes or writes better essays, now the question stands differently: whose model will more reliably protect the banking sector, cloud services, or government systems.

Altman is betting on security, and $10 million is just the first installment in a game where stakes will only grow with each new release. The key takeaway: OpenAI is transitioning from creating "smart assistants" to developing foundational protection tools. Will GPT-5.

3-Codex become the gold standard for secure development, or is this just another marketing shield that will shatter upon its first collision with reality?