Image: John Tredennick, Merlin Search Technologies with AI.
Our exploration of generative AI in trial preparation has demonstrated how Large Language Models can analyze complex materials to generate sophisticated closing arguments. We began by showing how to overcome traditional LLM length limitations, then demonstrated these techniques by creating closing arguments for both TransOcean and Halliburton in the BP Deepwater Horizon trial. Now we take on perhaps the most challenging perspective: the government’s case against all three defendants.
The Context
The government’s position in the Deepwater Horizon trial presented unique challenges for our AI system. Unlike individual defendants who focused on minimizing their own liability, the Department of Justice needed to present a comprehensive case demonstrating how the collective actions of BP, TransOcean, and Halliburton led to disaster. This required synthesizing evidence of both individual culpability and systematic failures across multiple organizations.
The DOJ’s case characterized the Gulf oil spill as an example of “gross negligence and willful misconduct,” citing a “culture of corporate recklessness” supported by internal communications showing BP officials were aware of risks but proceeded recklessly. The government also accused TransOcean, the rig’s owner and operator, of gross negligence.
The government’s strategy proved effective. U.S. District Judge Carl Barbier ultimately ruled that BP was guilty of gross negligence and willful misconduct, describing their actions as “reckless,” while finding TransOcean’s and Halliburton’s actions “negligent.” His ruling apportioned 67% of blame to BP, 30% to TransOcean, and 3% to Halliburton. This exposed BP to potential Clean Water Act fines of up to $18 billion, while TransOcean and Halliburton faced lesser penalties due to their lower degree of negligence.
Leveraging GenAI for the Government’s Case
Using our DiscoveryPartner platform, we processed the same trial materials used for the defense arguments but modified our approach to support the government’s position. This required our AI to:
- Identify evidence of negligence across all defendants
- Find interconnections between different parties’ actions
- Build a narrative showing how individual failures combined to create disaster
- Support arguments for joint and several liability
This broader analytical scope demonstrates another key advantage of GenAI in trial preparation: the ability to quickly analyze vast amounts of evidence from different strategic angles while maintaining a coherent narrative thread.
We did all of this work in less than two hours (closer to an hour really).
John Tredennick and Dr. William Webber, Merlin Search Technologies.
This broader analytical scope demonstrates another key advantage of GenAI in trial preparation: the ability to quickly analyze vast amounts of evidence from different strategic angles while maintaining a coherent narrative thread.
We did all of this work in less than two hours (closer to an hour really). Imagine the time it might take for a team of associates to create a first draft like this. Weeks at a minimum. Here is how we did it.
Building the Government’s Closing Argument
Our first step was creating an outline focused on systemic failures and collective responsibility. The AI-generated outline provided a framework for demonstrating how multiple parties’ negligent actions combined to cause the catastrophe:
This initial outline demonstrates how AI can help organize complex arguments involving multiple parties. While individual defendants could focus on minimizing their own liability, the government needed to show how each party’s actions contributed to an interconnected chain of failures.
The outline’s emphasis on technical failures, communication breakdowns, and operational shortcuts reflects the government’s strategy of demonstrating that no single error caused the disaster. Rather, it resulted from a systemic failure of corporate responsibility across multiple organizations.
Building the Argument Section by Section
After establishing our outline, we systematically built each section of the closing argument using DiscoveryPartner’s AI capabilities. Three complementary AI approaches were used to analyze the trial record:
- Semantic search to identify conceptually related content
- AI-enhanced keyword analysis to find specific technical details
- Machine learning classification to recognize patterns of evidence
For each section of the outline, these combined approaches identified approximately 700 relevant document sections. The AI then evaluated these materials for both factual support and strategic value to the government’s position.
What makes this approach particularly powerful for complex multi-defendant cases is how it synthesizes information across multiple sources. The AI identifies patterns and connections that support the government’s theory of collective responsibility while maintaining precise links to the underlying evidence.
The Government’s Argument
To demonstrate how this works in practice, here are the opening sections of the government’s closing argument. While this represents only an initial draft, it shows how AI can synthesize evidence from multiple sources into a coherent narrative about collective responsibility.
To demonstrate how this works in practice, here are the opening sections of the government’s closing argument. While this represents only an initial draft, it shows how AI can synthesize evidence from multiple sources into a coherent narrative about collective responsibility.
John Tredennick and Dr. William Webber, Merlin Search Technologies.
You can read and download the entire closing argument here.
This approach differs significantly from our previous defense-oriented arguments. Rather than focusing on a single party’s limited role or another’s primary responsibility, the government’s closing needed to weave together evidence of multiple failures while maintaining a clear narrative about collective responsibility.
Rather than focusing on a single party’s limited role or another’s primary responsibility, the government’s closing needed to weave together evidence of multiple failures while maintaining a clear narrative about collective responsibility.
John Tredennick and Dr. William Webber, Merlin Search Technologies.
Key Features of the AI-Generated Argument
Several aspects of this AI-generated argument deserve attention:
- Systematic Integration of Evidence
- Each major point is supported by specific document references
- Evidence from multiple defendants is woven together to show systematic failures
- Technical details are presented clearly without losing their complexity
- Narrative Structure
- Opens with broad context about systemic failures
- Progresses logically through specific technical and operational issues
- Maintains focus on collective responsibility while detailing individual failures
- Strategic Elements
- Emphasizes interconnections between different parties’ actions
- Builds case for joint and several liability through factual presentation
- Addresses both specific incidents and broader patterns of negligence
Creating the Complete Argument
The final step involved combining all sections into a complete closing argument. This process is straightforward using standard document tools, but requires attention to:
- Maintaining consistent formatting across sections
- Ensuring smooth transitions between major points
- Preserving all evidence citations
- Checking narrative flow and argument progression
The complete argument demonstrates how AI can help construct a comprehensive case involving multiple defendants while maintaining precise connections to the underlying evidence.
The Benefits of an AI Approach
This example highlights several key advantages of using AI for complex trial preparation:
- Comprehensive Analysis
- Processes thousands of pages of trial materials in minutes
- Identifies relevant evidence across multiple parties and issues
- Maintains precise tracking of supporting documentation
- Strategic Flexibility
- Generates arguments from different strategic perspectives
- Adapts analysis to support different legal theories
- Helps identify connections that might be missed in manual review
- Efficiency Gains
- Produces sophisticated first drafts quickly
- Allows legal teams to focus on strategy and refinement
- Reduces time spent on initial document review and organization
- Quality Control
- Every point is linked to specific evidence
- Arguments are built on comprehensive document analysis
- Helps ensure no key evidence is overlooked
Ultimately, the fact that one person, who has no basic knowledge of the case other than from the news years ago, could use a GenAI-enabled discovery platform to create what is essentially a good first draft leaves me speechless. The power of these GenAI tools to analyze, synthesize and report is something we have never even imagined until now.
The Path Forward
Our success in generating closing arguments from multiple perspectives in the BP trial demonstrates the practical value of AI in complex litigation. This technology is ready for deployment in real-world cases, offering legal teams powerful tools for:
- Exploring alternative theories of liability
- Testing arguments from different perspectives
- Ensuring comprehensive use of available evidence
- Accelerating trial preparation while maintaining quality
The key lies in combining AI’s analytical capabilities with human legal expertise. The technology doesn’t replace professional judgment – it enhances it by allowing lawyers to work more efficiently with complex evidence while maintaining rigorous analytical standards.
The future of legal practice increasingly depends on thoughtful integration of artificial and human intelligence. Tools like those demonstrated here show how technology can amplify legal expertise rather than replace it, leading to better outcomes for clients and the justice system.