MiniMax reports its unreleased MIM model solved 94 TII McEliece challenge instances, 55 previously unsolved
MiniMax published research on November 20 describing cryptanalysis results from MIM, an unreleased internal system. The company says MIM solved 94 TII McEliece challenge instances, including 55 it describes as previously unsolved, and states the work does not affect deployed cryptographic standards.
MiniMax says its unreleased internal research system, MiniMax's Internal Model (MIM), solved 94 instances in the TII McEliece challenges, 55 of which the company describes as previously unsolved. The results come from an English-language research blog and a public repository overview published on November 20. All findings are reported by MiniMax and have not been independently replicated.
What MiniMax reports
The company groups its results into three areas:
- Analysis of past NIST competition candidates
- Examination of peer-reviewed cryptographic constructions
- Faster solvers for mathematical challenge problems
Beyond the McEliece figures, MiniMax reports improved analysis of HAWK challenge parameters and weaknesses in selected earlier designs. It claims practical results and speed improvements over prior analyses. Those comparisons depend on the selected targets, parameters, hardware and reference methods, and the performance advantages have not been independently verified.
MiniMax presents the experiments as a test of sustained mathematical reasoning combined with implementation work.
What the results do not claim
MiniMax explicitly says the work does not affect deployed cryptographic standards. The three categories should not be read as a claim that an AI system has broken encryption protecting everyday communications.
- Historical candidates submitted to a standardization competition are not the same as the standards ultimately selected.
- Public challenge instances test particular parameters, not every deployment of a related cryptosystem.
- The 94 solved instances are challenge instances. They are not recovered keys from live services, and they do not amount to a compromise of an entire standardized algorithm family.
Disclosure and artifacts
MiniMax says it shared its findings with the respective scheme authors in advance. This is a company statement. It does not confirm that every author has endorsed the findings or completed any necessary remediation.
The repository overview lists technical reports and supporting proof-of-concept artifacts released on November 20. These are research artifacts, not open model weights.
Access to MIM
MIM is not a public model release. MiniMax is offering testing access by application to qualified researchers. Parameter count, context window, pricing and training details have not been disclosed.
What this means
The disclosure is a research report, not a product launch, and the company is careful about its scope. The headline figures, 94 solved McEliece challenge instances with 55 described as previously unsolved, concern public challenge parameters. They say nothing on their own about deployed systems, and MiniMax states that deployed standards are unaffected.
The claims remain company-reported. The speed comparisons depend on targets, parameters, hardware and baseline methods, so independent cryptographic review is needed to establish their scope and significance. Access limited to qualified researchers by application is one route for that kind of external scrutiny, alongside the technical reports and proof-of-concept artifacts MiniMax published. Until reviewers weigh in, the results are best read as a documented set of claims about specific historical candidates, academic designs and challenge instances.
Related Articles
Salesforce's BLIP-Live video agent cuts unnecessary responses by over 50%, company claims
Salesforce AI Research introduced BLIP-Live, a general model for live video agents that decides when to speak and when to stay silent. According to Salesforce, in one evaluation across 326 ego-centric videos it reduced unnecessary responses of state-of-the-art models by over 50%.
Anthropic: Selective notes beat full logs for Claude agent memory, with 'Dreaming' fixing errors
Anthropic published research on how Claude retains knowledge across sessions using note files. Its tests show selective notes outperform exhaustive archives, and a post-session process called Dreaming can correct faulty notes. All results come from Anthropic's own tests.
OpenAI outlines four-part approach to combat adversarial distillation
OpenAI describes a layered approach to adversarial distillation. It combines model training, safeguards, detection, and enforcement to protect against capability extraction.
Anthropic mathematician Levent Alpöge claims Claude-assisted proof of the Riemann hypothesis, posts it in Lean
Levent Alpöge, a mathematician employed by Anthropic, says he has proven the Riemann hypothesis, first proposed in 1859, working with Claude. He posted the proof coded in Lean, and experts have yet to confirm that it is correct.
Comments
Loading...