Now reinforces your safety while GPT -5 appears on the horizon, according to ft

The irruption of Deepseek In the market it did not go unnoticed. On January 20, 2025 he put his Free chatbotand in a matter of days climbed up to the top of the ranking in the American App Store, surpassing Chatgpt. The impact was such that it even affected directly to the stock market value of NVIDIA.

This success soon generated friction. Towards the end of that same month, Openai launched a serious accusation: the Chinese startup would have used its closed models to train its open source alternative. In statements to Financial Timesthe company directed by Sam Altman He claimed to have evidence of a distillation process that pointed to his Chinese rival.

OpenAI reinforces your security measures

Since then almost six months have passed. And if something has become clear, it was not allowed to pass the episode overlooking. According to the aforementioned British newspaperthe company initiated a thorough review of its internal practices and reinforced their security measures to reduce the risk of leaks or what internally they already refer as “corporate espionage.”

“The episode motivated Openai to be much more rigorous,” said someone close to the security team. The American company has intensified its efforts to shield inside: it has not only expanded its cybersecurity workforce, it has also reviewed its protocols and hardened its internal policies.

Among the most relevant measures is the isolation of much of its patented technology, which is now managed in out -of -line environments or separated from the rest of its networks. It has also implemented biometric controls of access to your offices, reinforced the physical security of your data centers and applies a policy of “Denegated output by default“To protect the weights of their models.

If you wonder what this last measure consists of, the answer is quite simple: it implies that any critical data transfer is blocked by general norm, and it is only allowed if it has been expressly authorized. Thus, any attempt to extract the models – or sensitive parts of them – is automatically denied unless it is manually enabled, minimizing the risk of escape.

This armor comes at a key moment for the company. The career for leadership in artificial intelligence has accelerated, and Openai is no longer seen as an unattainable force. The competition narrowstalent is expensive –Goal has begun paying exorbitant figures by their engineers– And all eyes are put on the arrival of GPT –5.

Although OpenAi has not confirmed the exact date, Sam Altman has said that The new model is on its way and that will be “a great evolution” with respect to what we know. Everything indicates that GPT -5 will unify the best of the previous versionswith greater reasoning capacity, better customization, and, presumably, a broader context window. The pressure to launch it without errors, in an increasingly competitive environment, is maximum.

Images | Solen Feyissa | Xataka with Grok

In Xataka | Huawei and Alibaba’s grudge threatens to transform into a real war: that of AI between Chinese companies