Claude Mythos model scores 100% on cyber scale:Anthropic created powerful AI but stopped its launch due to fear of cyber catastrophe

AI giant Anthropic on Wednesday introduced the world’s most powerful model, ‘Claude Mythos’, but it has not been launched for general users. The company says that this model works responsibly on its own, but its power could be extremely dangerous in the wrong hands. Banks, hospitals, power grids, and government systems could all become its targets. Anthropic has currently provided it to 12 large companies like Amazon, Google, Microsoft, and over 40 organizations under Project Glasswing, so that they can first improve the security of their systems. The name Project Glasswing is taken from a butterfly with transparent wings. It remains invisible even when in the open, just as software flaws remain hidden. Anthropic has promised ?8.5 billion in credit and ?340 million in aid to open-source organizations. After 90 days, the company will announce how many flaws the model has detected. Historic – A flaw not caught in 5 million tests, it found it on the first try In the pilot, the model discovered thousands of serious cyber vulnerabilities in many operating systems and web browsers. It found a 27-year-old vulnerability in OpenBSD, the most secure operating system. A 16-year-old vulnerability in video software FFmpeg, which was not found even after 5 million scans. It caught it on the first try. This model also showed a way to gain control of the entire system by adding several vulnerabilities to the Linux kernel. Benchmark – 100% score in cyber testing, 93.9% in coding This model achieved 100% success in the most difficult cybersecurity tests. In coding benchmarks, it scored 93.9%, while the previous model’s score was 80.8%. In CyberGym tests, it scored 83.1%, compared to the previous model’s 66.6%. It is worth noting that with the launch of this model, Anthropic’s annual revenue has also more than tripled to approximately 2.5 trillion rupees. CEO Amodei said – The Mythos model poses a great threat We created Mythos not for cybersecurity, but for coding. But the side effect of coding’s power was that it became equally proficient in cyber attacks. If it remains in the right hands, this technology can create a more secure internet and world than ever before. But if it goes wrong, the danger is immense.’