Anthropic’s Claude Mythos, deemed too powerful for public release, now appears to have new capabilities.
On Wednesday, the UK AI Security Institute (AISI) reported that a new version of Mythos outperformed its earlier iterations and OpenAI’s GPT-5.5, just a month after its initial release.
“The latest Mythos Preview checkpoint completed both cyber ranges, solving ‘The Last Ones’ in 6 of 10 attempts and the previously unsolved ‘Cooling Tower’ in 3 of 10 attempts,” the blog authors noted.
This was the first time a model completed our second cyber range. Anthropic’s announcement of Mythos Preview and Project Glasswing, a cybersecurity testing alliance with other tech companies and AI labs, received evaluation from UK AISI.
They concluded that the model is a significant improvement over previous frontier models in the evolving landscape of cyber performance.”
The third-party perspective balanced claims about Mythos, indicating that the excitement was more than just marketing but not a drastic leap in AI capabilities. The truth likely lies somewhere in between.
AISI’s updated test shows that improvements in AI capabilities can occur within a single model’s versions, not just in new releases. AI models are rapidly advancing in cyber task performance, greatly impacting cybersecurity, especially with Mythos’ effectiveness in detecting software vulnerabilities.
“In February 2026, we estimated that AI models could complete cyber tasks in half the time, doubling their capacity every 4.7 months since late 2024, compared to our previous estimate of 8 months in November 2025.”
AISI has reported two new models, Claude Mythos Preview and OpenAI’s GPT-5.5, which have greatly surpassed previous doubling rates. It is still unclear whether this trend will persist.
Mythos and GPT-5.5 may mark a significant shift in model evolution, but AISI acknowledged unknown factors in its testing. By capping tasks at 2.5 million tokens, researchers were able to make better performance comparisons over time.
The authors argued that the current description “underestimates what frontier models can achieve.” They highlighted that “Mythos Preview and GPT-5.5 show large error margins due to nearly 100% success rates on our longest tasks, despite the 2.5 million token limit.” They noted that their tasks are insufficiently long to assess reliability over longer durations.
The latest models are challenging the limits of our test suite, making it hard to identify failure points. This also suggests that success rates would be significantly higher without the token cap, potentially rendering time-horizon calculations nearly impossible.
Models with access to more tokens and advanced infrastructure are significantly more capable. The blog stated, “A 2.5 million token limit is low. In our cyber range experiment, we use up to 100 million tokens, and performance likely improves further with more tokens, especially for newer models.”
Luke 21:11 And great earthquakes shall be in divers places, and famines, and diseases; and fearful sights and great signs shall there be from heaven.
Luke 21:25 And there shall be signs in the sun, and in the moon, and in the stars; and upon the earth distress of nations, with perplexity; the sea and the waves roaring;
Luke 21:28 And when these things begin to come to pass, then look up, and lift up your heads; for your redemption draweth nigh.
Revelations 14:9 And the third angel followed them, saying with a loud voice, If any man worship the beast and his image, and receive his mark in his forehead, or in his hand,
Revelations 14:10 The same shall drink of the wine of the wrath of God, which is poured out without mixture into the cup of his indignation; and he shall be tormented with fire and brimstone in the presence of the holy angels, and in the presence of the Lamb:
Revelations 14:11 And the smoke of their torment ascendeth up for ever and ever: and they have no rest day nor night, who worship the beast and his image, and whosoever receiveth the mark of his name.
Revelations 14:12 Here is the patience of the saints: here are they that keep the commandments of God, and the faith of Jesus.
Revelations 14:13 And I heard a voice from heaven saying unto me, Write, Blessed are the dead which die in the Lord from henceforth: Yea, saith the Spirit, that they may rest from their labours; and their works do follow them.
2 Peter 3:10 But the day of the Lord will come as a thief in the night; in the which the heavens shall pass away with a great noise, and the (elements) shall melt with fervent heat, the earth also and the works that are therein shall be burned up.
Read more at: AI Has Already Become a Master of Lies And Deception
Read more at: Is the”Image” of the Beast in Revelations 13 “Alive”? (transformedbythetruth.com)
Read more at: https://transformedbythetruth.com/c-e-r-n/
Read more at: Former Google Engineer Issues Grave Warning About AI being “ALIVE” (transformedbythetruth.com)
Read more at: Google and Oxford scientists claims AI will soon destroy mankind (transformedbythetruth.com)
Read more at: Apocalyptic Events Ushering in the End Of The World? (transformedbythetruth.com)
Click here to read more articles transformedbythetruth.com