从需求到证明:望安科技完成AI驱动的大规模代码自动形式化验证 From requirements to proof: Wonsec completes an AI-driven, end-to-end formal verification practice for code

蓝绿色发光形式化验证单体与圆角科技分区地面概念图
pre(s) ∧ execute(s, ops, s′, out)
head = tail = 2⁶⁴ − 1
out = fifo(ops)
2⁶⁴ mod 3 = 1
Γ ⊢ spec ⟹ safe(code)
187 条items 独立验证需求获得明确结论Independent requirements reached definitive conclusions
140 条items 需求经证明在约定前提下成立Requirements proven valid under agreed assumptions
47 条items 需求由 Isabelle 核验的反例证明不成立Requirements disproved by Isabelle-verified counterexamples

产品动态 Product updates

AI时代痛点与解决AI-era challenges and solutions

AI 加速研发,也放大了可信性挑战AI accelerates development—and amplifies trust challenges

生成效率快速提升之后,需求准确性、结果可信度、验证覆盖与工程追溯成为新的关键问题。As generation accelerates, requirement accuracy, result trustworthiness, verification coverage and engineering traceability become critical.

需求理解易产生偏差Requirement meaning drifts in implementation

同一句需求,业务和研发可能有不同理解。代码写出来后,偏差才逐渐显现。The same requirement may be understood differently by business and engineering teams; discrepancies emerge only after the code is produced.

同一需求在业务理解与研发实现之间产生偏差的示意图
望安解法Wonsec solution

让关键需求有明确的验收标准Give critical requirements clear acceptance criteria

代码生成快,质量确认慢Generation outruns quality assurance

AI 缩短了开发时间,团队仍需确认生成的代码能否满足关键功能和安全要求。AI shortens development time, but teams must still confirm the generated code meets critical functional and security requirements.

AI 生成代码后仍需确认功能与安全质量的示意图
望安解法Wonsec solution

为关键功能提供正确性证据Provide correctness evidence for critical functions

少见风险易被遗漏Rare risks escape routine tests

有些问题只在特定操作顺序或长期运行后出现,常用测试场景很难覆盖。Some defects appear only under specific operation sequences or long-running conditions that routine tests rarely cover.

常规测试通过但边界风险仍可能遗漏的示意图
望安解法Wonsec solution

在产品交付前识别边界风险Identify boundary risks before product delivery

版本迭代,结论难沿用New versions invalidate old conclusions

需求和代码更新后,团队往往难以判断哪些结论仍然有效、哪些功能需要重新验证。When requirements and code change, teams struggle to know which conclusions remain valid and what must be verified again.

版本持续迭代导致原有结论失效并需要重新验证的示意图
望安解法Wonsec solution

让验证证据随版本更新Keep verification evidence current with every version

AI + 形式化验证模式AI + Formal Verification Model

从软件需求出发,用可核验的证据回答:代码能否按预期运行Start from software requirements and use verifiable evidence to answer whether the code behaves as intended.

确认验证目标Confirm the verification target

工程师确认代码应在什么条件下满足哪项要求Engineers confirm which requirement the code should satisfy and under what conditions.

AI AGENT

协同生成Collaborative generation

  • 理解需求与标准Understand needs and standards将业务目标与适用规范转化为清晰的生成约束Turn business goals and applicable standards into clear generation constraints
  • 生成研发资产Generate engineering assets生成代码、文档与测试用例,并保持与目标一致Generate code, documents, and test cases while keeping them aligned with the target
  • 规划验证任务Plan verification tasks拆解形式化验证与合规检查任务,准备待核验输入Break down formal verification and compliance checks into review-ready inputs
提交候选证明Submit candidate proof
返回待证目标Return unresolved target
形式化验证Formal verification

独立核验Independent verification

  • 形式化建模与规约Formal modeling and specification将关键需求转化为可计算、可证明的数学模型Translate critical requirements into computable, provable mathematical models
  • 安全属性验证Security-property verification独立分析代码与模型是否满足既定安全性质Independently determine whether code and models satisfy defined security properties
  • 高等级要求核验High-assurance requirement checks以可复核证据确认高等级安全要求是否成立Use reviewable evidence to confirm whether high-assurance requirements hold
要求成立Requirement holds

给出证明及适用条件Provide the proof and applicable conditions

发现反例Counterexample found

给出触发条件及偏离预期的行为Show triggers and behavior that departs from expectations

AI + 形式化,让严格验证兼具全面高效、可复用、可追溯AI + formal methods make rigorous verification comprehensive, efficient, reusable and traceable

AI 辅助证明提效,形式化验证确保推导正确,让每一次的验证成果汇聚为可信、可复用的验证资产。AI accelerates proof engineering while formal verification ensures correct reasoning, turning every verification result into a trusted, reusable asset.

更强的保证Stronger assurance

机器可核验Machine-checkable
超越有限样例Beyond finite examples
覆盖极端边界Extreme-boundary coverage

更高的效率Higher efficiency

提升证明工程效率Improve proof-engineering efficiency
验证知识持续复用Continuously reuse verified knowledge

更完整的证据More complete evidence

同时回答满足与不满足Show both satisfaction and violation
端到端可追溯End-to-end traceability
可信验证资产Trusted verification assets 可检查Inspectable 可复核Reviewable 可复用Reusable

    安全合规服务Security compliance services

    一站式合规实施与 CertHub AI 协同,覆盖认证推进、证据管理与持续合规。One-stop delivery and CertHub AI connect certification, evidence management and continuous compliance.

    咨询与实施Consulting & delivery

    一站式安全合规服务One-stop compliance service

    专属合规 Fellow 统筹差距评估、材料准备、测评对接与整改闭环。A dedicated fellow coordinates assessment, evidence, evaluation and remediation.

    了解服务Explore
    ◎范围与路径Scope & path明确认证目标与实施路线Define targets and route
    ◇项目与协作CollaborationFellow 统筹多方协同推进Coordinate every stakeholder
    ▤材料与交付Evidence delivery材料准备与测评对接Prepare evidence and evaluation
    ✓整改与闭环Remediation差距修复直至认证交付Close gaps through certification
    AI 合规平台AI compliance platform

    CertHub AI 智能合规平台compliance platform

    AI 辅助标准解析、差距识别和证据生成,让合规过程可协作、可追溯。AI-assisted analysis, gap identification and evidence generation make compliance traceable.

    查看平台View platform
    AI标准智能解析Standard analysis快速理解条款与适用要求Interpret clauses and scope
    △差距识别Gap discovery定位差异并形成改进任务Find gaps and create tasks
    ∞证据版本关联Evidence linkage持续关联材料、版本与要求Link evidence and versions
    ↻在线评审闭环Review workflow协同评审与反馈持续闭环Collaborate and close reviews