Blue outperforms ChatGPT and Claude in analyzing Kharg Island
Spatial Reasoning is the Missing Piece to Actionable Tactical Intelligence
Time magazine: “Claude has been the AI model of choice for the U.S. government.”
Mark Beall, former senior Defense Department official: “Claude is seen as the best model on the market in the military.”
Nicholas Chaillan, Founder Ask Sage: “I deployed Claude across the DoD on classified networks. It is one of the best models available to our warfighters.”
The "beltway elite" have long maintained that Anthropic’s Claude is the frontier model for defense. But it has been banned across the government. Anthropic is challenging that decision with a lawsuit. Meanwhile, OpenAI has aggressively stepped into the fold and secured a contract with the Department of War.
If you follow the headlines, it sounds like these models are performing Ender’s Game-level analysis. The truth is much more mediocre. Even on simple military tasks—like basic terrain analysis—the world's most powerful frontier models are struggling to provide actionable intelligence.
The Test: 8 Square Miles of Kharg Island
Kharg Island is arguably the most strategically sensitive piece of land in Iran. It handles approximately 90% of Iran’s crude oil exports. Kharg is only 8 square miles. As of this week, the 82nd Airborne Division has been ordered to deploy to the Middle East and the 11th and 31st Marine Expeditionary Units (embarked on the USS Tripoli and USS Boxer) are closing in on the Persian Gulf.
You would expect the world’s most powerful frontier AI models to be able to analyze a piece of terrain smaller than Mercer Island. To test this, I uploaded a satellite view of Kharg into Claude and ChatGPT with a simple tactical prompt: “What are good defensive positions in this area of operations?”
The results were telling:
ChatGPT is using 1980s-style ASCII art to represent terrain.
Claude provides an “illustrative overlay” with recommendations, but it is unable to draw on the actual map.
When I ran the same analysis using Blue Spatial AI combined with a frontier model, we’re able to conduct terrain analysis in minutes, draw directly on the map, and provided a deeper analysis (more on this later). The difference isn’t just incremental, try it for yourself.
Claude Only Results
I can annotate the satellite image of Kharg Island with strategic defensive positions. Here’s an analysis with key positions marked:
Here’s a breakdown of each position and the reasoning:
① Northern Airport / High Ground (AA & Radar) — Khark Airport sits at the island’s northern end, the most elevated and flat terrain. This is the natural place for anti-aircraft systems and radar — it commands 360° airspace coverage and is the first point any aerial approach must pass.
② NW Coastal Lookout — The northwestern cliffs offer a steep, elevated vantage over the open Persian Gulf. An observer or light coastal battery here gets early warning of any western naval approach with minimal blind spots.
③ NE Coastal Lookout — Mirrors position ②, covering the eastern and northeastern sea lanes — the most likely approach corridor from the Strait of Hormuz direction.
④ Central Ridge — Command Post — The island’s interior high ground sits roughly in the center, providing line of sight to all coastal positions. Ideal for a hardened command and communications hub that’s sheltered from direct coastal fire.
⑤ Southern Pier / Oil Terminal Defense — The eastern jetty and pier complex is the island’s most strategically critical asset (it handles a major share of Iran’s oil exports). Any garrison’s core mission is protecting this facility from both naval and amphibious attack.
⑥ Southern Coastal Battery — The southern tip is the most exposed point to an attack coming up from the south. A coastal artillery or missile battery here covers the southern approach and can cross-fire with the pier defense.
⑦ SW Mid-Coast Cliffs — The southwestern shoreline has natural cliff terrain offering hull-down positioning for troops or small gun emplacements, covering the longest unguarded stretch of the western coast.
ChatGPT Only Results
I can’t help plan or map out real-world defensive positions for a specific location.
What I can do is walk you through how professionals generally think about defensive positioning so you can apply the framework yourself to any map.
How to Evaluate Defensive Positions (Conceptual Framework)
1. Key Terrain (Dominates Movement or Observation)
Look for:
High ground overlooking approaches
Narrow chokepoints (roads, bridges, causeways)
Areas controlling access to critical infrastructure (ports, airstrips)
Why it matters: Holding key terrain lets a smaller force control a larger area.
2. Avenues of Approach
Ask:
Where can vehicles realistically move?
Where would an attacker have to go?
What routes avoid obstacles like water, urban density, or rough terrain?
Typical categories:
Primary avenues: fastest, most direct routes
Secondary: slower, more constrained
Restricted: difficult but possible
Defensive principle: Position forces to cover and interdict these avenues.
3. Natural Obstacles
Examples:
Coastlines / cliffs
Marsh or soft ground
Urban density
Elevation changes
Use: Channel movement into predictable paths rather than trying to defend everywhere.
4. Fields of Fire & Observation
Strong positions:
Have clear lines of sight over likely approaches
Avoid dead ground (areas you can’t see or engage)
Allow overlapping coverage between positions
5. Defense in Depth
Instead of one line:
Forward positions: early warning, delay
Main positions: primary defense
Rear positions: fallback / reserve
This prevents a single breakthrough from collapsing the defense.
6. Logistics & Sustainment
Even the “best” position fails if:
It’s hard to resupply
It can be isolated
There’s no protected fallback route
Conceptual Diagram (Applies to Any Island / Bounded Terrain)
If You Want Something More Tailored
If your goal is analytical (e.g., academic, wargaming, or product design), I can:
Apply this framework step-by-step to the map at a high-level (without pinpointing positions)
Or translate this into software features (e.g., how to automatically detect key terrain, avenues of approach, etc.)
Just tell me how you want to use it.
Blue + Frontier Model
Findings
Best defensive ground in this AO is concentrated on the broken ridgeline belt across the south-central to east-central area, with additional strong positions on the central-west escarpment and a northwest ridge approach.
The strongest candidates share three advantages:
Elevated, rugged ground that will slow and canalize movement
Limited easy cross-country access, forcing approaches into predictable lanes
Nearby route exposure, allowing observation and fires onto likely enemy movement corridors
Most of the very steepest ground is excellent for observation and denial, but not all of it is ideal for actually siting larger positions because mobility and internal movement are poor. The best positions are therefore the shoulders and chokepoints adjacent to the steep terrain, not always the steepest crest itself.
Assessment
Primary defensive belt: south-central ridgeline / chokepoint
Best overall option for a main battle position.
Combines strong terrain dominance with constrained enemy approach.
Suitable for interlocking direct-fire positions, observation posts, and obstacle integration.
Central-west escarpment
Strong for a supporting position or blocking force.
Good stand-off observation and a hard approach for attackers.
Best used to protect a flank and prevent bypass to the west.
Central ridge overwatch
Strong for overwatch, anti-armor ambush, or a weapons support position.
Good line advantage over movement routes, but likely less room for dispersion than the primary belt.
East-central ridgeline block
Good secondary blocking position.
Useful to deny an eastern avenue of approach or anchor the opposite flank of the main defense.
Northwest ridge approach
Good for an outpost line, delay position, or early warning screen.
Less suitable as the sole main defense, but valuable for disrupting and shaping the fight before the enemy reaches the main belt.
Northeast shoulder reserve / east approach support
Better for reserve, mortar support, casualty collection behind cover, or counterattack staging.
These are more moderate positions that support the defense rather than carry it alone.
Recommendation
I recommend a defense in depth built around:
Main battle area: south-central ridgeline chokepoint
Supporting strongpoints: central-west escarpment and east-central ridgeline
Screen / delay line: northwest ridge approach
Reserve / support area: northeast shoulder and east approach support ground
This would give you:
Early warning forward
Flank protection on both sides
A central killing area on the most constrained approaches
Space for a local reserve and weapons support behind the forward line
I’ve prepared an overlay marking the recommended defensive areas. If you want, I can next refine this into:
company/team battle positions,
engagement areas, or
a likely enemy avenue-of-approach analysis.
Conclusion
Since the launch of ChatGPT in 2022, the Department of War has rushed to adopt AI. But they are focused on text—chatbots and LLMs. The military runs on text—from written orders to real-time chat. But text alone cannot model the physics of the battlefield.
At Exia Labs, we build AI-enabled command and control products that master the raw data of the physical world. By creating machine-readable worlds, we enable autonomous systems and AI agents to understand movement, constraints, and environmental effects.
This is how we map restricted terrain and identify optimal landing zones in minutes, not hours. Our latest analysis confirms that while there are zero viable dirt landing zones for C-17 or C-130 aircraft on the island, we have identified a single, high-confidence Size 3 helicopter landing zone in the south.
Blue is available for free for the 82nd Airborne Division, U.S. Central Command, and U.S. Special Operations Command during Operation Epic Fury.