King's Indian Attack vs the French Defense and the Caro-Kann: the setup, the plan, and what the engine says

Short answer

The King's Indian Attack builds the same structure every game — Nf3, d3, g3, Bg2, Nbd2 and short castling — and only then picks a plan. This guide walks the setup move by move against the French, the Caro-Kann and the Sicilian, explains the e4-e5 advance that closes the centre and moves the game to the kingside, and describes the Nd2-f1-e3 manoeuvre behind the f4-f5 break. It also measures all three base positions in Stockfish 18: every one of them level, and why that is not an argument against the system.

The essentials in a few lines

• The King's Indian Attack does not claim an opening advantage. In the base position against the French, Stockfish 18 reads +0.03 at depth 28 — dead level. What the system saves is decisions, not evaluation.
• The move that turns the setup into an attack is e4-e5: it closes the centre, frees e4 for the d2 knight and sends the game to the wing where White already has bishop, knight and queen available.
• Against the Caro-Kann the system yields least, and the reason is arithmetic: Black spends a move on ...c6 and still reaches ...e5 in one go, occupying the centre the KIA wanted to attack. The reading drops to −0.21.
• Black's counterplay is always the same and comes from the far side of the board: ...b5, ...b4 and pressure down the c-file. The game becomes a race between two attacks on opposite wings, and the side whose king is the target cannot afford a slow move.
• In Fischer–Myagmarsuren (Sousse, 1967) the engine reads move 17 as 0.00 and the game ends in mate sixteen moves later. The evaluation was not wrong: it measured the position, not how hard it was to defend.

A system, not a variation

The difference between a system and a theoretical opening is where the knowledge lives. In a theoretical opening it lives in the sequence: move 9 depends on the opponent's move 8, and getting the order wrong costs the position. In a system it lives in the final structure — the pieces go to the same squares, and the order adapts to what the opponent does. The King's Indian Attack is the purest example of that inside a 1.e4 repertoire.

The name comes from the mirror image: it is the King's Indian Defence played by White, with an extra tempo. The g2 fianchetto, the pawn on d3 and the knight on d2 form the same strongpoint Black builds in the King's Indian — and because White moves first, the manoeuvre that arrives late in the defence arrives on time here. That is why the system survives being played against defences with nothing in common.

In repertoire terms this solves a concrete problem: answering the French, the Caro-Kann and the Sicilian with a single body of knowledge. Each of those has a main line with its own theory and hundreds of catalogued moves; against the KIA they all land in a similar structure, and the difference between them becomes a matter of plan rather than memorised variation. What that economy costs is in the next section, and it can be measured.

What the engine says about the system

It is worth facing the number head-on, because it is the main argument of everyone who advises against the system. Against the French, with the setup complete and Black developed naturally, the position reads +0.03: nothing. Against the Sicilian, 0.00. Against the Caro-Kann in Black's best version, −0.21. None of those positions is bad, and none is better than what 3.Nc3 gets you against the French.

The right reading of that set is not «the system is no good» but «the system does not buy an advantage — it buys something else». What it buys is asymmetry of preparation: White knows the plan for the next fifteen moves, and Black, who studied twenty moves of Winawer or Classical Caro-Kann, arrives at move 8 in a position that was not part of the file. The edge is in clock time and familiarity, and neither of those shows up on the evaluation bar.

There is a real cost hidden in the same data, and it deserves stating plainly: against an opponent who knows the system, the KIA holds no surprise at all. Black knows the attack comes from the right, knows the answer is ...b5-b4 on the left, and the game turns into a technical race won by whoever understands the structure better. The system is a shortcut around ignorance, not around preparation.

e4-e5 and the knight manoeuvre

Before e4-e5 the setup is merely sensible development with no pretensions. After it, the position acquires a whole logic: the e5 pawn cuts the g2 bishop's diagonal but fixes Black's structure, the d2 knight — now useless where it stands — gains the f1-e3 route, and g5 opens for the other knight. The attack does not exist before that move. That move is what creates the subject.

The Nd2-f1 manoeuvre is the detail that separates players who understood the system from players who copied it. The knight looks like it is retreating, and it is: it abandons the defence of the centre precisely because the centre is closed and no longer needs defending. From f1 it goes to e3, eyeing d5 and f5, or to g3, supporting h4-h5. Leaving that knight on d2 «for safety» leaves you attacking with two pieces, and two is not enough.

The final break is usually h4-h5 followed by f4-f5, or the two in reverse order depending on how Black defends. The target is not a piece: it is the square that opens when the h-pawn or the f-pawn disappears — h6, g6 or f6, according to Black's setup. It is worth naming the requirement: enough pieces in the sector before the file opens. Opening early, with two attackers and the rest of the army on the other wing, is the classic first-time KIA mistake.

Against the French, where it pays most

The reason the French is the KIA's best customer is structural. The e6 pawn is already committed when White plays e4-e5, which means Black has spent a move locking in the c8 bishop and then receives the advance that fixes the lock. Black's worst piece in the French is still the worst piece here, with the added problem that no open file arrives to compensate for it.

Black's counterplay is well known and entirely legitimate: ...b7-b5, ...b5-b4 and pressure on the c-file, backed by the c5 pawn already in place. The game becomes a race between two attacks. What usually decides it is the quality of the target: White attacks a king, Black attacks pawns and squares. When both sides move at the same speed, the side aiming at the king has the practical edge — and it is that asymmetry, not the evaluation, that keeps the system alive.

One Black reply deserves attention because it changes the game: an early ...d5-d4, closing the centre on Black's terms and taking space before White does. The engine reads that line at +0.60 for White, which makes it less frightening than it looks — but White's plan has to change with it. With Black's d4 pawn fixed, the flank attack loses urgency and the move becomes c2-c3, undermining the head of the chain.

• 2.d3 dispenses with Winawer, Tarrasch and Advance theory in one move.
• e4-e5 arrives with tempo on the f6 knight, almost always on move 8 or 9.
• The c8 bishop, already poor in the French, has no improvement plan left after e5.
• Black attacks with ...b5-b4 and the c-file; White with h4-h5 and f4-f5.
• Against an early ...d4, change targets: c2-c3 against the base of the chain, not the flank attack.

Against the Caro-Kann, where it pays least

The comparison with the French makes the point. In the French, Black's pawn reaches e5 in two steps and White occupies the square first; in the Caro-Kann, ...c6 blocks nothing and ...e5 comes straight away. The result is a symmetrical, open structure — the exact opposite of what the system wants. With no closed centre, the Nf1-e3 manoeuvre loses its point and the flank attack has nothing behind it.

The measurement confirms the impression and puts a size on it. With ...e5 played, Stockfish 18 gives −0.21 at depth 28; when Black instead develops with ...g6 and leaves the centre alone, the same engine gives +0.30. It is the largest gap this guide records between two Black choices in one opening — half a pawn separating two lines that differ by a single move.

The practical consequence belongs to repertoire, not to theory: anyone playing the KIA for convenience needs either a second answer to the Caro-Kann or the willingness to accept equality and play an ordinary middlegame. Neither is a problem. The problem is building the setup expecting the usual attack and discovering on move 10 that there is no advance to play.

When not to play it, and how to train it

The practical rule, for building a repertoire, is to look at Black's e-pawn. If it stands on e6, the system has a target; if it goes to e5 on move one, it has none; if it can reach e5 in a single move, as in the Caro-Kann, the target depends on Black cooperating. That one question settles the decision before move 3 and stops you playing the system out of habit in positions that contradict it.

Training the KIA differs from training a theoretical opening, and the difference is what makes it cheap. There is no line to memorise: there are three typical positions to recognise — before e5, after e5 and after the f4-f5 break — and a handful of manoeuvres that keep recurring. An afternoon reviewing games in the system, watching for the moment each one opened the wing, teaches more than any list of variations.

One warning the model game hides. Fischer beat Myagmarsuren cleanly, and the game circulates as proof that the system wins by itself; the position at move 17 was rated 0.00. What decided it was not the opening. It was the difference between the player who knew where he was going and the player who had to work it out. Replaying Fischer's moves without the plan reproduces the structure and not the result.

Frequently asked questions

Openings mentioned in this guide

• King's Indian Attack (A07)
• French Defense (C00)
• Caro-Kann Defense (B10)
• Réti Opening (A04)