Anthropic's Claude found weaknesses in real cryptography, including a scheme designed to survive quantum computers.
In research published July 28 with partners at ETH Zurich, Tel Aviv University and TU Berlin, Claude halved the effective key size of HAWK-256, a post-quantum signature candidate under NIST review, and improved a classic attack on reduced-round AES by up to 800 times. Each attack cost roughly $100,000 of compute and took days, not decades.
@anthropic-inc is straight about what this is not: HAWK is not deployed, the AES variant was weakened for research, and nothing you use today is broken. What changed is who does the cryptanalysis. Flaw-finding that once belonged to a small community of specialists ran as a model working autonomously for about three days.
Gerald Mullally, CEO of @oxfordquantumcircuits set out the standard threat model on our show: 'Classical compute will take 2 trillion years to break encryption', while a quantum machine needs hours. Then he asked the question this research has now answered: 'how many of our problems actually require brute force?'
Every post-quantum migration plan assumes the danger arrives with hardware, on a hardware timeline. This result points at a nearer risk, and a nearer defence: cognition. An attacker does not need a 2-trillion-year search if a model can find the structure that makes the search unnecessary. The question for a security budget stops being how big your adversary's computer is, and becomes how good their cryptanalyst is, and whether that cryptanalyst rents for $100,000 a run.
If AI can find the cracks before quantum can force the door, which belongs in your threat model first?
Sources:
Anthropic's research: https://www.anthropic.com/research/discovering-cryptographic-weaknesses · Full conversation with Gerald Mullally, CEO of Oxford Quantum Circuits: https://youtu.be/FjY1_1PGykY
#Cybersecurity #QuantumComputing #AI
In research published July 28 with partners at ETH Zurich, Tel Aviv University and TU Berlin, Claude halved the effective key size of HAWK-256, a post-quantum signature candidate under NIST review, and improved a classic attack on reduced-round AES by up to 800 times. Each attack cost roughly $100,000 of compute and took days, not decades.
@anthropic-inc is straight about what this is not: HAWK is not deployed, the AES variant was weakened for research, and nothing you use today is broken. What changed is who does the cryptanalysis. Flaw-finding that once belonged to a small community of specialists ran as a model working autonomously for about three days.
Gerald Mullally, CEO of @oxfordquantumcircuits set out the standard threat model on our show: 'Classical compute will take 2 trillion years to break encryption', while a quantum machine needs hours. Then he asked the question this research has now answered: 'how many of our problems actually require brute force?'
Every post-quantum migration plan assumes the danger arrives with hardware, on a hardware timeline. This result points at a nearer risk, and a nearer defence: cognition. An attacker does not need a 2-trillion-year search if a model can find the structure that makes the search unnecessary. The question for a security budget stops being how big your adversary's computer is, and becomes how good their cryptanalyst is, and whether that cryptanalyst rents for $100,000 a run.
If AI can find the cracks before quantum can force the door, which belongs in your threat model first?
Sources:
Anthropic's research: https://www.anthropic.com/research/discovering-cryptographic-weaknesses · Full conversation with Gerald Mullally, CEO of Oxford Quantum Circuits: https://youtu.be/FjY1_1PGykY
#Cybersecurity #QuantumComputing #AI
- Категория
- Заработок на AI
Выполните вход или зарегистрируйтесь для отправки комментария.
Комментариев нет.
