Skip to content
Submit incident
AII-1000-37611
ConfirmedA court, a regulator or the organization itself has stated the core facts, or the record cites a court or regulator source.

Anthropic's Mythos AI discovers vulnerabilities in encryption standards

Anthropic's AI model Mythos identified weaknesses in a weakened version of the Advanced Encryption Standard (AES) during research, reportedly 200 to 1,000 times faster than human experts, and also found security flaws in the HAWK next-generation encryption algorithm.

28 July 2026 Β· Anthropic Β· Security and cyberattacks
28 July 2026
AII-1000-376112 sources ↓
Quick facts
Category
Security and cyberattacks
Country
United States
Source
ν•œκ΅­μΌλ³΄
Date
28 July 2026
Drafted by
Raidu Incident Detection

What happened

On July 28, 2026, Anthropic announced that its AI model Mythos preview version successfully identified vulnerabilities in a weakened version of the Advanced Encryption Standard (AES) during a research experiment.

The Mythos model discovered security flaws in the HAWK algorithm, a next-generation encryption standard being considered by the U.S. National Institute of Standards and Technology (NIST) for the quantum computing era.

Mythos identified these vulnerabilities 200 to 1,000 times faster than human cryptography experts.

Reported impact

Anthropic stated the research findings have no impact on current internet service operations, as the experiment involved an intentionally weakened AES version.

Researchers noted that large language models were previously thought incapable of surpassing humans in cryptography, but Mythos now performs advanced cryptographic research not yet discovered by the academic community.

Index analysis

Anthropic researcher Nicholas Carini stated that one year prior, AI could not solve cryptography problems he could solve at age 16, but now performs research-level cryptography work.

The New York Times reported that the discovery challenges the prior understanding that large language models could not surpass humans in cryptography.

Governance controls

Anthropic stated it expects AI models using advanced cryptographic review to serve as tools for developing stronger encryption standards and safer internet security protocols.

The U.S. Trump administration imposed export controls on Anthropic's Mythos5 and Fable5 models in June 2026 due to cyber terrorism concerns, suspending international services through early July.

Classification

Category
Security and cyberattacks
Industry
Digital security
Country
United States
Harm status
Potential harm
Intent
Unintentional
Timing
Post-deployment
Harm types
Public interest
Consequences
None recorded
AI systems
Mythos

Sources and evidence

This record was researched and written by the Index, based on the following original reporting.

ν•œκ΅­μΌλ³΄
"λ―Έν† μŠ€, 인터넷 μ•”ν˜Έ 체계 무λ ₯ν™” κ°€λŠ₯μ„±"... μ»€μ§€λŠ” 'AI ν•΄ν‚Ή' μœ„ν—˜-κ΅­μ œγ…£ν•œκ΅­μΌλ³΄
29 July 2026
μ—°ν•©λ‰΄μŠ€
NYT "λ―Έν† μŠ€, 인터넷 μ•”ν˜Έμ²΄κ³„ 무λ ₯ν™” κ°€λŠ₯μ„±"
29 July 2026