Inside Anycase 2.0: Building the most accurate legal AI in the Philippines

Anycase 2.0: How we improved our system’s accuracy 

The Anycase 2.0 update introduces key AI upgrades that greatly improve how it identifies and cites legal references, while filtering out misleading or off-topic citations for higher accuracy and relevance. 

These improvements were tested on 120 questions from the 2024 Philippine Bar Exams, comparing Anycase 2.0, Anycase 1.0, and recent GPT models on the same core competencies expected of every Filipino lawyer.

Legal accuracy and actionability of answers 

Anycase 2.0 sets a new standard for legal reliability, with 87.5% of its answers rated accurate, defensible, and immediately-usable in a lawyer’s output. The upgrade marks a clear leap from Anycase 1.0, cutting incorrect conclusions by half.

GPT-5 performed well in general reasoning, but around 8% of its outputs still contained serious legal inaccuracies. 

0% irrelevance rate 

In the same benchmark, Anycase 2.0 completely eliminated the 1.7% irrelevance rate recorded in Anycase 1.0, demonstrating a major improvement in the system’s ability to identify and retrieve only the most accurate and relevant legal references.

Active fabrication and hallucinations

Anycase 2.0 maintained a 0% fabrication rate in the same evaluations using 120 questions from the 2024 Philippine Bar Exams. Each answer was traceable to a real case or established doctrine, ensuring every citation pointed to a legitimate source. 

In contrast, 32.5% of GPT with Web Search’s answers contained fabricated or unsupported claims. 

What next: Deepening legal coverage and verifiability

With Anycase 2.0, accuracy is no longer the challenge, but the foundation. The system now delivers consistently accurate answers, grounded in real, verifiable references. 

From here, our goal is to ensure that lawyers not only get correct answers but also see exactly where those answers come from, and know that every relevant law and case has been surfaced.

Verifiability of citations and claims 

Anycase 2.0 maintained the same high standard of verifiability established in earlier versions, with 94.2% of its answers fully traceable to legitimate legal sources, compared to 85.8% for Anycase 1.0. 

In comparison, GPT with Web Search continued to trail behind, with only 55.8% of answers verifiable, with more than half of its answers lacking citations or traceable legal references. 

Improved legal coverage 

Anycase 2.0 returned complete legal coverage in 75% of its answers, marking a clear improvement from Anycase 1.0’s 52.5%. Meanwhile, ChatGPT with Web Search missed key citations and legal authorities in more than half of its answers. 

Anycase 2.0: Our most accurate AI system yet  

Each round of evaluations shows how Anycase’s accuracy continues to strengthen. With accuracy now at a stable point, our focus moves to what comes next: verifiability and legal coverage.

• Verifiability: How can lawyers confirm where an answer came from? We’re building clearer source trails that show exactly which sections of a case, law, or rule informed the response.

• Legal coverage: Did Anycase retrieve all the authorities needed to start analysis? We’re improving Anycase’s ability to provide the complete scope of laws, doctrines, and jurisprudence relevant to the query.

These updates will make every answer fully traceable, giving lawyers a clear view of how the AI forms its analysis.

Lawyers using Anycase already save an average of 14 hours a week on research. As our verifiability and legal coverage metrics improve, those hours will go even further.

Core Anycase 2.0 improvements 

Upgraded legal AI 

Anycase 2.0 marks our first comprehensive upgrade to our legal AI systems. The new system minimizes irrelevant citations and confusing context, surfacing only the most relevant legal references to help you find clear answers faster.

9,800+ new data sources

The strategic addition of the following government documents directly improves the accuracy and completeness of Anycase’s answers for related queries: 

We expanded coverage across key government agencies to ensure more complete, authoritative answers:

• DOLE Department Orders, Labor Advisories, and Department Circulars

• NLRC En Banc Resolutions, Notable Decisions, and Rules of Procedure 

• BIR Revenue Regulations, Memorandum Circulars, Revenue Regulations, Revenue Memorandum Orders, Revenue Memorandum Circulars, Revenue Administrative Orders, and Revenue Delegation of Authority Orders

• 2024 DOJ-NPS Rules on Preliminary Investigations and Inquest Proceedings (Department Circular No. 15, series of 2024)

More direct, helpful answers 

Anycase 2.0’s upgraded AI delivers clearer, more focused responses. Each answer now addresses your query right from the start, while still grounding its explanation in the relevant legal basis. It’s also far less likely to ignore your question or return a vague response.

Built against hallucination 

Guarding against AI hallucinations has always been central to how we build Anycase. Every answer includes citations, grounding responses in real and verifiable legal authorities.

Each citation redirects users to the full text when clicked. Anycase.ai’s AI answer engine is built on a centralized legal corpus that scans government websites daily for new updates and uploads.

Anycase 2.0: A more helpful legal AI for Filipino law students 

ALAC replaces IRAC in Edu Mode

Our core team includes both practicing legal professionals and law students who continuously test Anycase.ai’s performance for Bar readiness. Edu Mode now follows the ALAC method (Answer, Law, Application, Conclusion) providing clearer, more direct explanations of legal doctrines than the previous IRAC structure.

Improved digest with case trail and comprehensive facts 

When we noticed “digest” queries getting frequent downvotes by law students and practitioners alike, we dug deeper and rebuilt how Anycase answers digest-related queries. Case digests now include:

• More comprehensive statement of facts

• Dedicated sections summarizing each party’s arguments

• Case trail tracing rulings from the lower courts up to the Supreme Court

• Fuller ratio decidendi, with all key issues clearly identified and discussed

Suggested examples: 

• Digest Duterte v. HOR (2025)

• Digest GR 160756, emphasize the discussion on MCIT

• Digest Edu v. Ericta, only discuss the issue on police power

Built for bar-type questions and evaluations 

The new chat UI lets law students ask Anycase for Bar-type questions in paragraph format, mirroring the structure of actual exam prompts. After writing your own answer, you can ask Anycase to evaluate it, checking for completeness, accuracy, and alignment with established doctrines.

Here are some queries to try: 

• “Can you give me a Bar-type question in paragraph form about the rules on evidence, focusing on the admissibility of object evidence?” 

  • “Here is my answer to the object evidence question—can you evaluate it and suggest improvements?”

• “Give me 10 bar-type questions on Tax Law” 

• “Give me a paragraph-form Bar question that tests the latest Supreme Court doctrine on psychological incapacity.”

How to access Anycase 2.0 (Standard v2)? 

Anycase 2.0 is now available to Free, Pro, and Edu users as the new default AI answer mode.
To access Edu Mode (ALAC), simply switch to Edu in your settings. Learn more about how to change Anycase.ai’s answer modes here.