Artificial intelligence news.
Straight from the original announcements.

SciGround Adds Mathematical Verification Layer for LLM Research

November 13, 2026, 14:43By the AI News Bites Editorial TeamSource: PR Newswire
SciGround Adds Mathematical Verification Layer for LLM Research
Image: PR Newswire

Licklider, Inc. has launched SciGround, a mathematical verification layer that checks whether an LLM-driven research project’s design, data, analysis and claim agree. The company says researchers can continue using their preferred AI assistant for reasoning while SciGround performs verification outside the model.

Key points
  • Licklider launched SciGround on Nov. 13, 2026.
  • The system checks research design, data, analysis and claims outside the LLM.
  • SciGround MCP public beta, v0.1.0, connects MCP-capable assistants.
  • Results include machine-readable evidence, assumptions, limits and software versions.
  • The launch supports a defined set of statistical and research workflows.

Verification moves outside the AI assistant

SciGround is designed for research workflows in which large language models produce hypotheses, analysis code and manuscript text. Licklider says the ability to verify that work has not kept pace with the scale of AI-generated research.

The company points to a July 2026 report from the White House Office of Science and Technology Policy, which stated that “the cost of generation has decreased exponentially, the cost of verification has not.” It also cites a September editorial in Nature Computational Science that described reproducibility with LLMs as “arguably more challenging.”

Rather than replacing the researcher’s AI assistant, SciGround works alongside it. The assistant calls SciGround through the Model Context Protocol, or MCP, while calculations and checks run in established software under recorded versions.

Fixed methods and machine-readable evidence

SciGround fixes the test, comparison direction and threshold before results are assessed. If the design or data do not fit a supported workflow, the system refuses or pauses instead of approximating. Licklider also says it does not switch methods after seeing significance.

The checks cover the full path from design to claim. SciGround examines whether the data match the design, whether the numbers hold within a stated error bound and whether the claim follows from the result.

Calculations and checks use software including SciPy, NumPy, R, FLINT/Arb for guaranteed-precision arithmetic and the Z3 solver. Each result is returned as a machine-readable Record containing the claim, assumptions, limits, evidence, inputs and software versions. When data are corrected, affected results and claims are rechecked, with evidence retained for each version.

Initial workflows and MCP access

SciGround’s supported workflows at launch include Student's and Welch's t-tests, paired t-tests matched by ID, one-way ANOVA with pre-specified corrected comparisons, the Mann, Whitney U test, and several outcomes with a pre-specified correction.

The launch list also includes bulk RNA-seq differential expression with FDR control, equipment-failure missing data, three fixed batches, original-versus-replication comparison and a two-look interim design. Licklider says each workflow has defined inputs and claims.

SciGround MCP is available as a public beta, version 0.1.0, and connects the verification layer to any assistant that supports MCP. “An LLM is an excellent partner to think with,” Licklider founder and CEO Tasuku Kobayashi said. “But the proof, not the model, should speak for a conclusion. SciGround makes AI-assisted research something others can check.”

Licklider, Inc.SciGroundSciGround MCPModel Context ProtocolscienceAI safetyenterprise AI

Related news

Biointron launches RushDATA platform for AI antibody discovery
Products and toolsNovember 13, 2026, 19:22

Biointron launches RushDATA platform for AI antibody discovery

Biointron has launched RushDATA™, an integrated wet-lab validation platform for AI- and machine learning-driven antibody discovery. The company says it can deliver…

Products and tools
Products and toolsNovember 12, 2026, 19:21

RemotePC Adds MCP Integration for AI-Assisted Computer Management

RemotePC has introduced an integration that lets customers use ChatGPT, Claude and other compatible AI assistants to monitor remote computers and perform supported…

Anaconda’s Enkrypt AI named Gartner sample vendor for multimodal protection
Safety and ethicsNovember 4, 2026, 15:50

Anaconda’s Enkrypt AI named Gartner sample vendor for multimodal protection

Enkrypt AI, now part of Anaconda, has been named a Sample Vendor for Multimodal AI Protection in Gartner’s Emerging Tech Impact Radar: AI Cybersecurity Ecosystem. The…