INTO is a one-byte instruction (CEh) that looks at OF. With OF = 0 it does nothing and execution goes on. With OF = 1 it acts exactly like INT 4: push FLAGS, clear IF and TF, push CS, push IP, and continue at the address stored in vector 4 (0000:0010). The saved IP points to the instruction after the INTO, so a handler that only reports the error can return with IRET and the program continues.

The idea is to put INTO after each signed operation and handle every overflow in one place instead of writing a JO everywhere:

ADD     AX, BX
INTO  ; INT 4 if the signed sum overflowed
IMUL    CX
INTO  ; and if the product needs DX

It only helps once a program has installed its own INT 4 handler (with DOS function 25h, restoring the old vector before exit). What the vector points to otherwise depends on the system, so the runnable example makes the same test with JO, without an interrupt:

Prints "signed overflow". Change 10 to 5 and it prints "no overflow".
MOV     AL, 120
ADD     AL, 10  ; 130 does not fit a signed byte: OF = 1
MOV     DX, OFFSET OKMSG1
JNO     SHOW1   ; the test INTO makes
MOV     DX, OFFSET OVMSG1
SHOW1:
MOV     AH, 09h
INT     21h
 
OKMSG1  DB      "no overflow", 13, 10, "$"
OVMSG1  DB      "signed overflow", 13, 10, "$"

Later CPUs

INTO works the same on later x86 processors in 16- and 32-bit code; 64-bit mode dropped it.