diff --git a/src/dsp/dsp.h b/src/dsp/dsp.h
index 57006e22..f1e3365b 100644
--- a/src/dsp/dsp.h
+++ b/src/dsp/dsp.h
@@ -19,8 +19,11 @@
 #endif
 
 #include "src/dsp/cpu.h"
+#include "src/utils/bounds_safety.h"
 #include "src/webp/types.h"
 
+WEBP_ASSUME_UNSAFE_INDEXABLE_ABI
+
 #ifdef __cplusplus
 extern "C" {
 #endif
diff --git a/src/utils/bit_reader_inl_utils.h b/src/utils/bit_reader_inl_utils.h
index 4374538e..2db0d825 100644
--- a/src/utils/bit_reader_inl_utils.h
+++ b/src/utils/bit_reader_inl_utils.h
@@ -26,10 +26,13 @@
 #include "src/dsp/cpu.h"
 #include "src/dsp/dsp.h"
 #include "src/utils/bit_reader_utils.h"
+#include "src/utils/bounds_safety.h"
 #include "src/utils/endian_inl_utils.h"
 #include "src/utils/utils.h"
 #include "src/webp/types.h"
 
+WEBP_ASSUME_UNSAFE_INDEXABLE_ABI
+
 #ifdef __cplusplus
 extern "C" {
 #endif
@@ -79,7 +82,7 @@ static WEBP_UBSAN_IGNORE_UNDEF WEBP_INLINE void VP8LoadNewBytes(
         : "memory", "at");
 #else
     lbit_t in_bits;
-    memcpy(&in_bits, br->buf, sizeof(in_bits));
+    WEBP_UNSAFE_MEMCPY(&in_bits, br->buf, sizeof(in_bits));
 #endif
     br->buf += BITS >> 3;
 #if !defined(WORDS_BIGENDIAN)
diff --git a/src/utils/bit_reader_utils.c b/src/utils/bit_reader_utils.c
index f1042304..b34a84aa 100644
--- a/src/utils/bit_reader_utils.c
+++ b/src/utils/bit_reader_utils.c
@@ -21,10 +21,13 @@
 #include "src/dsp/cpu.h"
 #include "src/utils/bit_reader_inl_utils.h"
 #include "src/utils/bit_reader_utils.h"
+#include "src/utils/bounds_safety.h"
 #include "src/utils/endian_inl_utils.h"
 #include "src/utils/utils.h"
 #include "src/webp/types.h"
 
+WEBP_ASSUME_UNSAFE_INDEXABLE_ABI
+
 //------------------------------------------------------------------------------
 // VP8BitReader
 
@@ -258,7 +261,7 @@ static void PrintBitTraces(void) {
 void BitTrace(const struct VP8BitReader* const br, const char label[]) {
   int i, pos;
   if (!init_done) {
-    memset(kLabels, 0, sizeof(kLabels));
+    WEBP_UNSAFE_MEMSET(kLabels, 0, sizeof(kLabels));
     atexit(PrintBitTraces);
     buf_start = br->buf;
     init_done = 1;
diff --git a/src/utils/bit_reader_utils.h b/src/utils/bit_reader_utils.h
index 9c5d849f..f926dcf7 100644
--- a/src/utils/bit_reader_utils.h
+++ b/src/utils/bit_reader_utils.h
@@ -22,8 +22,11 @@
 #include <stdlib.h>  // _byteswap_ulong
 #endif
 #include "src/dsp/cpu.h"
+#include "src/utils/bounds_safety.h"
 #include "src/webp/types.h"
 
+WEBP_ASSUME_UNSAFE_INDEXABLE_ABI
+
 // Warning! This macro triggers quite some MACRO wizardry around func signature!
 #if !defined(BITTRACE)
 #define BITTRACE 0  // 0 = off, 1 = print bits, 2 = print bytes
diff --git a/src/utils/bit_writer_utils.c b/src/utils/bit_writer_utils.c
index 54076cb1..bd2355a1 100644
--- a/src/utils/bit_writer_utils.c
+++ b/src/utils/bit_writer_utils.c
@@ -18,10 +18,13 @@
 #include <stdlib.h>
 #include <string.h>  // for memcpy()
 
+#include "src/utils/bounds_safety.h"
 #include "src/utils/endian_inl_utils.h"
 #include "src/utils/utils.h"
 #include "src/webp/types.h"
 
+WEBP_ASSUME_UNSAFE_INDEXABLE_ABI
+
 //------------------------------------------------------------------------------
 // VP8BitWriter
 
@@ -46,7 +49,7 @@ static int BitWriterResize(VP8BitWriter* const bw, size_t extra_size) {
   }
   if (bw->pos > 0) {
     assert(bw->buf != NULL);
-    memcpy(new_buf, bw->buf, bw->pos);
+    WEBP_UNSAFE_MEMCPY(new_buf, bw->buf, bw->pos);
   }
   WebPSafeFree(bw->buf);
   bw->buf = new_buf;
@@ -180,7 +183,7 @@ int VP8BitWriterAppend(VP8BitWriter* const bw, const uint8_t* data,
   assert(data != NULL);
   if (bw->nb_bits != -8) return 0;  // Flush() must have been called
   if (!BitWriterResize(bw, size)) return 0;
-  memcpy(bw->buf + bw->pos, data, size);
+  WEBP_UNSAFE_MEMCPY(bw->buf + bw->pos, data, size);
   bw->pos += size;
   return 1;
 }
@@ -188,7 +191,7 @@ int VP8BitWriterAppend(VP8BitWriter* const bw, const uint8_t* data,
 void VP8BitWriterWipeOut(VP8BitWriter* const bw) {
   if (bw != NULL) {
     WebPSafeFree(bw->buf);
-    memset(bw, 0, sizeof(*bw));
+    WEBP_UNSAFE_MEMSET(bw, 0, sizeof(*bw));
   }
 }
 
@@ -222,7 +225,7 @@ static int VP8LBitWriterResize(VP8LBitWriter* const bw, size_t extra_size) {
     return 0;
   }
   if (current_size > 0) {
-    memcpy(allocated_buf, bw->buf, current_size);
+    WEBP_UNSAFE_MEMCPY(allocated_buf, bw->buf, current_size);
   }
   WebPSafeFree(bw->buf);
   bw->buf = allocated_buf;
@@ -232,7 +235,7 @@ static int VP8LBitWriterResize(VP8LBitWriter* const bw, size_t extra_size) {
 }
 
 int VP8LBitWriterInit(VP8LBitWriter* const bw, size_t expected_size) {
-  memset(bw, 0, sizeof(*bw));
+  WEBP_UNSAFE_MEMSET(bw, 0, sizeof(*bw));
   return VP8LBitWriterResize(bw, expected_size);
 }
 
@@ -241,7 +244,7 @@ int VP8LBitWriterClone(const VP8LBitWriter* const src,
   const size_t current_size = src->cur - src->buf;
   assert(src->cur >= src->buf && src->cur <= src->end);
   if (!VP8LBitWriterResize(dst, current_size)) return 0;
-  memcpy(dst->buf, src->buf, current_size);
+  WEBP_UNSAFE_MEMCPY(dst->buf, src->buf, current_size);
   dst->bits = src->bits;
   dst->used = src->used;
   dst->error = src->error;
@@ -252,7 +255,7 @@ int VP8LBitWriterClone(const VP8LBitWriter* const src,
 void VP8LBitWriterWipeOut(VP8LBitWriter* const bw) {
   if (bw != NULL) {
     WebPSafeFree(bw->buf);
-    memset(bw, 0, sizeof(*bw));
+    WEBP_UNSAFE_MEMSET(bw, 0, sizeof(*bw));
   }
 }
 
diff --git a/src/utils/bit_writer_utils.h b/src/utils/bit_writer_utils.h
index 4a4bdb05..d0cdd01d 100644
--- a/src/utils/bit_writer_utils.h
+++ b/src/utils/bit_writer_utils.h
@@ -16,8 +16,11 @@
 
 #include <stddef.h>
 
+#include "src/utils/bounds_safety.h"
 #include "src/webp/types.h"
 
+WEBP_ASSUME_UNSAFE_INDEXABLE_ABI
+
 #ifdef __cplusplus
 extern "C" {
 #endif
diff --git a/src/utils/color_cache_utils.c b/src/utils/color_cache_utils.c
index eef08067..56f5353b 100644
--- a/src/utils/color_cache_utils.c
+++ b/src/utils/color_cache_utils.c
@@ -17,9 +17,12 @@
 #include <stdlib.h>
 #include <string.h>
 
+#include "src/utils/bounds_safety.h"
 #include "src/utils/utils.h"
 #include "src/webp/types.h"
 
+WEBP_ASSUME_UNSAFE_INDEXABLE_ABI
+
 //------------------------------------------------------------------------------
 // VP8LColorCache.
 
@@ -47,6 +50,6 @@ void VP8LColorCacheCopy(const VP8LColorCache* const src,
   assert(src != NULL);
   assert(dst != NULL);
   assert(src->hash_bits == dst->hash_bits);
-  memcpy(dst->colors, src->colors,
-         ((size_t)1u << dst->hash_bits) * sizeof(*dst->colors));
+  WEBP_UNSAFE_MEMCPY(dst->colors, src->colors,
+                     ((size_t)1u << dst->hash_bits) * sizeof(*dst->colors));
 }
diff --git a/src/utils/color_cache_utils.h b/src/utils/color_cache_utils.h
index efcab340..d85203f6 100644
--- a/src/utils/color_cache_utils.h
+++ b/src/utils/color_cache_utils.h
@@ -19,8 +19,11 @@
 
 #include "src/dsp/cpu.h"
 #include "src/dsp/dsp.h"
+#include "src/utils/bounds_safety.h"
 #include "src/webp/types.h"
 
+WEBP_ASSUME_UNSAFE_INDEXABLE_ABI
+
 #ifdef __cplusplus
 extern "C" {
 #endif
diff --git a/src/utils/endian_inl_utils.h b/src/utils/endian_inl_utils.h
index 9983e3eb..d561d640 100644
--- a/src/utils/endian_inl_utils.h
+++ b/src/utils/endian_inl_utils.h
@@ -17,8 +17,11 @@
 #endif
 
 #include "src/dsp/dsp.h"
+#include "src/utils/bounds_safety.h"
 #include "src/webp/types.h"
 
+WEBP_ASSUME_UNSAFE_INDEXABLE_ABI
+
 #if defined(WORDS_BIGENDIAN)
 #define HToLE32 BSwap32
 #define HToLE16 BSwap16
diff --git a/src/utils/filters_utils.c b/src/utils/filters_utils.c
index 9f93cbb0..39b81281 100644
--- a/src/utils/filters_utils.c
+++ b/src/utils/filters_utils.c
@@ -17,8 +17,11 @@
 #include <string.h>
 
 #include "src/dsp/dsp.h"
+#include "src/utils/bounds_safety.h"
 #include "src/webp/types.h"
 
+WEBP_ASSUME_UNSAFE_INDEXABLE_ABI
+
 // -----------------------------------------------------------------------------
 // Quick estimate of a potentially interesting filter mode to try.
 
@@ -34,7 +37,7 @@ WEBP_FILTER_TYPE WebPEstimateBestFilter(const uint8_t* data, int width,
                                         int height, int stride) {
   int i, j;
   int bins[WEBP_FILTER_LAST][SMAX];
-  memset(bins, 0, sizeof(bins));
+  WEBP_UNSAFE_MEMSET(bins, 0, sizeof(bins));
 
   // We only sample every other pixels. That's enough.
   for (j = 2; j < height - 1; j += 2) {
diff --git a/src/utils/filters_utils.h b/src/utils/filters_utils.h
index 70687540..c1e545ae 100644
--- a/src/utils/filters_utils.h
+++ b/src/utils/filters_utils.h
@@ -15,8 +15,11 @@
 #define WEBP_UTILS_FILTERS_UTILS_H_
 
 #include "src/dsp/dsp.h"
+#include "src/utils/bounds_safety.h"
 #include "src/webp/types.h"
 
+WEBP_ASSUME_UNSAFE_INDEXABLE_ABI
+
 #ifdef __cplusplus
 extern "C" {
 #endif
diff --git a/src/utils/huffman_encode_utils.c b/src/utils/huffman_encode_utils.c
index 5ce542b1..45f758ea 100644
--- a/src/utils/huffman_encode_utils.c
+++ b/src/utils/huffman_encode_utils.c
@@ -17,10 +17,13 @@
 #include <stdlib.h>
 #include <string.h>
 
+#include "src/utils/bounds_safety.h"
 #include "src/utils/utils.h"
 #include "src/webp/format_constants.h"
 #include "src/webp/types.h"
 
+WEBP_ASSUME_UNSAFE_INDEXABLE_ABI
+
 // -----------------------------------------------------------------------------
 // Util function to optimize the symbol map for RLE coding
 
@@ -228,7 +231,8 @@ static void GenerateOptimalTree(const uint32_t* const histogram,
               break;
             }
           }
-          memmove(tree + (k + 1), tree + k, (tree_size - k) * sizeof(*tree));
+          WEBP_UNSAFE_MEMMOVE(tree + (k + 1), tree + k,
+                              (tree_size - k) * sizeof(*tree));
           tree[k].total_count = count;
           tree[k].value = -1;
 
@@ -408,7 +412,7 @@ void VP8LCreateHuffmanTree(uint32_t* const histogram, int tree_depth_limit,
                            uint8_t* const buf_rle, HuffmanTree* const huff_tree,
                            HuffmanTreeCode* const huff_code) {
   const int num_symbols = huff_code->num_symbols;
-  memset(buf_rle, 0, num_symbols * sizeof(*buf_rle));
+  WEBP_UNSAFE_MEMSET(buf_rle, 0, num_symbols * sizeof(*buf_rle));
   OptimizeHuffmanForRle(num_symbols, buf_rle, histogram);
   GenerateOptimalTree(histogram, num_symbols, huff_tree, tree_depth_limit,
                       huff_code->code_lengths);
diff --git a/src/utils/huffman_encode_utils.h b/src/utils/huffman_encode_utils.h
index c1280476..98dc61da 100644
--- a/src/utils/huffman_encode_utils.h
+++ b/src/utils/huffman_encode_utils.h
@@ -14,8 +14,11 @@
 #ifndef WEBP_UTILS_HUFFMAN_ENCODE_UTILS_H_
 #define WEBP_UTILS_HUFFMAN_ENCODE_UTILS_H_
 
+#include "src/utils/bounds_safety.h"
 #include "src/webp/types.h"
 
+WEBP_ASSUME_UNSAFE_INDEXABLE_ABI
+
 #ifdef __cplusplus
 extern "C" {
 #endif
diff --git a/src/utils/huffman_utils.c b/src/utils/huffman_utils.c
index ee90a6d4..5b60a00e 100644
--- a/src/utils/huffman_utils.c
+++ b/src/utils/huffman_utils.c
@@ -17,10 +17,13 @@
 #include <stdlib.h>
 #include <string.h>
 
+#include "src/utils/bounds_safety.h"
 #include "src/utils/utils.h"
 #include "src/webp/format_constants.h"
 #include "src/webp/types.h"
 
+WEBP_ASSUME_UNSAFE_INDEXABLE_ABI
+
 // Huffman data read via DecodeImageStream is represented in two (red and green)
 // bytes.
 #define MAX_HTREE_GROUPS 0x10000
diff --git a/src/utils/huffman_utils.h b/src/utils/huffman_utils.h
index b8c71cde..1468b07b 100644
--- a/src/utils/huffman_utils.h
+++ b/src/utils/huffman_utils.h
@@ -16,9 +16,12 @@
 
 #include <assert.h>
 
+#include "src/utils/bounds_safety.h"
 #include "src/webp/format_constants.h"
 #include "src/webp/types.h"
 
+WEBP_ASSUME_UNSAFE_INDEXABLE_ABI
+
 #ifdef __cplusplus
 extern "C" {
 #endif
diff --git a/src/utils/palette.c b/src/utils/palette.c
index 6251db19..4b1c779d 100644
--- a/src/utils/palette.c
+++ b/src/utils/palette.c
@@ -18,12 +18,15 @@
 #include <string.h>
 
 #include "src/dsp/lossless_common.h"
+#include "src/utils/bounds_safety.h"
 #include "src/utils/color_cache_utils.h"
 #include "src/utils/utils.h"
 #include "src/webp/encode.h"
 #include "src/webp/format_constants.h"
 #include "src/webp/types.h"
 
+WEBP_ASSUME_UNSAFE_INDEXABLE_ABI
+
 // -----------------------------------------------------------------------------
 
 // Palette reordering for smaller sum of deltas (and for smaller storage).
@@ -82,7 +85,7 @@ int SearchColorNoIdx(const uint32_t sorted[], uint32_t color, int num_colors) {
 void PrepareMapToPalette(const uint32_t palette[], uint32_t num_colors,
                          uint32_t sorted[], uint32_t idx_map[]) {
   uint32_t i;
-  memcpy(sorted, palette, num_colors * sizeof(*sorted));
+  WEBP_UNSAFE_MEMCPY(sorted, palette, num_colors * sizeof(*sorted));
   qsort(sorted, num_colors, sizeof(*sorted), PaletteCompareColorsForQsort);
   for (i = 0; i < num_colors; ++i) {
     idx_map[SearchColorNoIdx(sorted, palette[i], num_colors)] = i;
@@ -188,7 +191,7 @@ static void PaletteSortMinimizeDeltas(const uint32_t* const palette_sorted,
                                       int num_colors, uint32_t* const palette) {
   uint32_t predict = 0x00000000;
   int i, k;
-  memcpy(palette, palette_sorted, num_colors * sizeof(*palette));
+  WEBP_UNSAFE_MEMCPY(palette, palette_sorted, num_colors * sizeof(*palette));
   if (!PaletteHasNonMonotonousDeltas(palette_sorted, num_colors)) return;
   // Find greedily always the closest color of the predicted color to minimize
   // deltas in the palette. This reduces storage needs since the
@@ -393,11 +396,12 @@ int PaletteSort(PaletteSorting method, const struct WebPPicture* const pic,
   switch (method) {
     case kSortedDefault:
       if (palette_sorted[0] == 0 && num_colors > 17) {
-        memcpy(palette, palette_sorted + 1,
-               (num_colors - 1) * sizeof(*palette_sorted));
+        WEBP_UNSAFE_MEMCPY(palette, palette_sorted + 1,
+                           (num_colors - 1) * sizeof(*palette_sorted));
         palette[num_colors - 1] = 0;
       } else {
-        memcpy(palette, palette_sorted, num_colors * sizeof(*palette));
+        WEBP_UNSAFE_MEMCPY(palette, palette_sorted,
+                           num_colors * sizeof(*palette));
       }
       return 1;
     case kMinimizeDelta:
diff --git a/src/utils/palette.h b/src/utils/palette.h
index 417c61fa..4f8971ab 100644
--- a/src/utils/palette.h
+++ b/src/utils/palette.h
@@ -14,8 +14,11 @@
 #ifndef WEBP_UTILS_PALETTE_H_
 #define WEBP_UTILS_PALETTE_H_
 
+#include "src/utils/bounds_safety.h"
 #include "src/webp/types.h"
 
+WEBP_ASSUME_UNSAFE_INDEXABLE_ABI
+
 struct WebPPicture;
 
 // The different ways a palette can be sorted.
diff --git a/src/utils/quant_levels_dec_utils.c b/src/utils/quant_levels_dec_utils.c
index 2d14eb9a..0dc53625 100644
--- a/src/utils/quant_levels_dec_utils.c
+++ b/src/utils/quant_levels_dec_utils.c
@@ -18,9 +18,12 @@
 
 #include <string.h>  // for memset
 
+#include "src/utils/bounds_safety.h"
 #include "src/utils/utils.h"
 #include "src/webp/types.h"
 
+WEBP_ASSUME_UNSAFE_INDEXABLE_ABI
+
 // #define USE_DITHERING   // uncomment to enable ordered dithering (not vital)
 
 #define FIX 16                            // fix-point precision for averaging
@@ -230,7 +233,7 @@ static int InitParams(uint8_t* const data, int width, int height, int stride,
   p->cur = p->start;
   p->end = p->start + R * width;
   p->top = p->end - width;
-  memset(p->top, 0, width * sizeof(*p->top));
+  WEBP_UNSAFE_MEMSET(p->top, 0, width * sizeof(*p->top));
   mem += size_scratch_m;
 
   p->average = (uint16_t*)mem;
@@ -270,7 +273,7 @@ int WebPDequantizeLevels(uint8_t* const data, int width, int height, int stride,
 
   if (radius > 0) {
     SmoothParams p;
-    memset(&p, 0, sizeof(p));
+    WEBP_UNSAFE_MEMSET(&p, 0, sizeof(p));
     if (!InitParams(data, width, height, stride, radius, &p)) return 0;
     if (p.num_levels > 2) {
       for (; p.row < p.height; ++p.row) {
diff --git a/src/utils/quant_levels_dec_utils.h b/src/utils/quant_levels_dec_utils.h
index ba4af446..3dcf4db6 100644
--- a/src/utils/quant_levels_dec_utils.h
+++ b/src/utils/quant_levels_dec_utils.h
@@ -14,8 +14,11 @@
 #ifndef WEBP_UTILS_QUANT_LEVELS_DEC_UTILS_H_
 #define WEBP_UTILS_QUANT_LEVELS_DEC_UTILS_H_
 
+#include "src/utils/bounds_safety.h"
 #include "src/webp/types.h"
 
+WEBP_ASSUME_UNSAFE_INDEXABLE_ABI
+
 #ifdef __cplusplus
 extern "C" {
 #endif
diff --git a/src/utils/quant_levels_utils.c b/src/utils/quant_levels_utils.c
index 3e4311f4..71401647 100644
--- a/src/utils/quant_levels_utils.c
+++ b/src/utils/quant_levels_utils.c
@@ -17,6 +17,7 @@
 #include <assert.h>
 #include <stddef.h>
 
+#include "src/utils/bounds_safety.h"
 #include "src/webp/types.h"
 
 #define NUM_SYMBOLS 256
@@ -24,6 +25,8 @@
 #define MAX_ITER 6            // Maximum number of convergence steps.
 #define ERROR_THRESHOLD 1e-4  // MSE stopping criterion.
 
+WEBP_ASSUME_UNSAFE_INDEXABLE_ABI
+
 // -----------------------------------------------------------------------------
 // Quantize levels.
 
diff --git a/src/utils/quant_levels_utils.h b/src/utils/quant_levels_utils.h
index 688f8546..43eecf2a 100644
--- a/src/utils/quant_levels_utils.h
+++ b/src/utils/quant_levels_utils.h
@@ -16,8 +16,11 @@
 
 #include <stdlib.h>
 
+#include "src/utils/bounds_safety.h"
 #include "src/webp/types.h"
 
+WEBP_ASSUME_UNSAFE_INDEXABLE_ABI
+
 #ifdef __cplusplus
 extern "C" {
 #endif
diff --git a/src/utils/random_utils.c b/src/utils/random_utils.c
index 63359873..24fe86c7 100644
--- a/src/utils/random_utils.c
+++ b/src/utils/random_utils.c
@@ -15,8 +15,11 @@
 
 #include <string.h>
 
+#include "src/utils/bounds_safety.h"
 #include "src/webp/types.h"
 
+WEBP_ASSUME_UNSAFE_INDEXABLE_ABI
+
 //------------------------------------------------------------------------------
 
 // 31b-range values
@@ -33,7 +36,7 @@ static const uint32_t kRandomTable[VP8_RANDOM_TABLE_SIZE] = {
     0x27e5ed3c};
 
 void VP8InitRandom(VP8Random* const rg, float dithering) {
-  memcpy(rg->tab, kRandomTable, sizeof(rg->tab));
+  WEBP_UNSAFE_MEMCPY(rg->tab, kRandomTable, sizeof(rg->tab));
   rg->index1 = 0;
   rg->index2 = 31;
   rg->amp = (dithering < 0.0) ? 0
diff --git a/src/utils/random_utils.h b/src/utils/random_utils.h
index 203723a9..1af9a62c 100644
--- a/src/utils/random_utils.h
+++ b/src/utils/random_utils.h
@@ -16,8 +16,11 @@
 
 #include <assert.h>
 
+#include "src/utils/bounds_safety.h"
 #include "src/webp/types.h"
 
+WEBP_ASSUME_UNSAFE_INDEXABLE_ABI
+
 #ifdef __cplusplus
 extern "C" {
 #endif
diff --git a/src/utils/rescaler_utils.c b/src/utils/rescaler_utils.c
index 114fa2ec..dc9ce676 100644
--- a/src/utils/rescaler_utils.c
+++ b/src/utils/rescaler_utils.c
@@ -19,9 +19,12 @@
 #include <string.h>
 
 #include "src/dsp/dsp.h"
+#include "src/utils/bounds_safety.h"
 #include "src/utils/utils.h"
 #include "src/webp/types.h"
 
+WEBP_ASSUME_UNSAFE_INDEXABLE_ABI
+
 //------------------------------------------------------------------------------
 
 int WebPRescalerInit(WebPRescaler* const rescaler, int src_width,
@@ -78,7 +81,7 @@ int WebPRescalerInit(WebPRescaler* const rescaler, int src_width,
   }
   rescaler->irow = work;
   rescaler->frow = work + num_channels * dst_width;
-  memset(work, 0, (size_t)total_size);
+  WEBP_UNSAFE_MEMSET(work, 0, (size_t)total_size);
 
   WebPRescalerDspInit();
   return 1;
diff --git a/src/utils/rescaler_utils.h b/src/utils/rescaler_utils.h
index d484d13a..259994b0 100644
--- a/src/utils/rescaler_utils.h
+++ b/src/utils/rescaler_utils.h
@@ -18,8 +18,11 @@
 extern "C" {
 #endif
 
+#include "src/utils/bounds_safety.h"
 #include "src/webp/types.h"
 
+WEBP_ASSUME_UNSAFE_INDEXABLE_ABI
+
 #define WEBP_RESCALER_RFIX 32  // fixed-point precision for multiplies
 #define WEBP_RESCALER_ONE (1ull << WEBP_RESCALER_RFIX)
 #define WEBP_RESCALER_FRAC(x, y) \
diff --git a/src/utils/thread_utils.c b/src/utils/thread_utils.c
index 184751f9..3dd34842 100644
--- a/src/utils/thread_utils.c
+++ b/src/utils/thread_utils.c
@@ -16,8 +16,11 @@
 #include <assert.h>
 #include <string.h>  // for memset()
 
+#include "src/utils/bounds_safety.h"
 #include "src/utils/utils.h"
 
+WEBP_ASSUME_UNSAFE_INDEXABLE_ABI
+
 #ifdef WEBP_USE_THREAD
 
 #if defined(_WIN32)
@@ -257,7 +260,7 @@ static void ChangeState(WebPWorker* const worker, WebPWorkerStatus new_status) {
 //------------------------------------------------------------------------------
 
 static void Init(WebPWorker* const worker) {
-  memset(worker, 0, sizeof(*worker));
+  WEBP_UNSAFE_MEMSET(worker, 0, sizeof(*worker));
   worker->status = NOT_OK;
 }
 
diff --git a/src/utils/thread_utils.h b/src/utils/thread_utils.h
index 815c944f..f9bc1727 100644
--- a/src/utils/thread_utils.h
+++ b/src/utils/thread_utils.h
@@ -18,8 +18,11 @@
 #include "src/webp/config.h"
 #endif
 
+#include "src/utils/bounds_safety.h"
 #include "src/webp/types.h"
 
+WEBP_ASSUME_UNSAFE_INDEXABLE_ABI
+
 #ifdef __cplusplus
 extern "C" {
 #endif
diff --git a/src/utils/utils.c b/src/utils/utils.c
index 04223e2c..d0e15258 100644
--- a/src/utils/utils.c
+++ b/src/utils/utils.c
@@ -17,10 +17,13 @@
 #include <stdlib.h>
 #include <string.h>  // for memcpy()
 
+#include "src/utils/bounds_safety.h"
 #include "src/utils/palette.h"
 #include "src/webp/encode.h"
 #include "src/webp/types.h"
 
+WEBP_ASSUME_UNSAFE_INDEXABLE_ABI
+
 // If PRINT_MEM_INFO is defined, extra info (like total memory used, number of
 // alloc/free etc) is printed. For debugging/tuning purpose only (it's slow,
 // and not multi-thread safe!).
@@ -228,9 +231,21 @@ void WebPSafeFree(void* const ptr) {
 
 // Public API functions.
 
-void* WebPMalloc(size_t size) { return WebPSafeMalloc(1, size); }
+void* WEBP_SINGLE WebPMalloc(size_t size) {
+  // Currently WebPMalloc/WebPFree are declared in src/webp/types.h, which does
+  // not include bounds_safety.h. As such, the "default" annotation for the
+  // pointers they accept/return is __single.
+  //
+  // All callers will need to immediately cast the returned pointer to
+  // WEBP_BIDI_INDEXABLE or WEBP_INDEXABLE via
+  // WEBP_UNSAFE_FORGE_BIDI_INDEXABLE.
+  //
+  // TODO: https://issues.webmproject.org/432511225 - Remove this once we can
+  // annotate WebPMalloc/WebPFree.
+  return WEBP_UNSAFE_FORGE_SINGLE(void*, WebPSafeMalloc(1, size));
+}
 
-void WebPFree(void* ptr) { WebPSafeFree(ptr); }
+void WebPFree(void* WEBP_SINGLE ptr) { WebPSafeFree(ptr); }
 
 //------------------------------------------------------------------------------
 
@@ -239,7 +254,7 @@ void WebPCopyPlane(const uint8_t* src, int src_stride, uint8_t* dst,
   assert(src != NULL && dst != NULL);
   assert(abs(src_stride) >= width && abs(dst_stride) >= width);
   while (height-- > 0) {
-    memcpy(dst, src, width);
+    WEBP_UNSAFE_MEMCPY(dst, src, width);
     src += src_stride;
     dst += dst_stride;
   }
diff --git a/src/utils/utils.h b/src/utils/utils.h
index 55c250df..bcb9119c 100644
--- a/src/utils/utils.h
+++ b/src/utils/utils.h
@@ -21,8 +21,11 @@
 
 #include <assert.h>
 
+#include "src/utils/bounds_safety.h"
 #include "src/webp/types.h"
 
+WEBP_ASSUME_UNSAFE_INDEXABLE_ABI
+
 #ifdef __cplusplus
 extern "C" {
 #endif
@@ -69,7 +72,7 @@ WEBP_EXTERN void WebPSafeFree(void* const ptr);
 // memcpy() is the safe way of moving potentially unaligned 32b memory.
 static WEBP_INLINE uint32_t WebPMemToUint32(const uint8_t* const ptr) {
   uint32_t A;
-  memcpy(&A, ptr, sizeof(A));
+  WEBP_UNSAFE_MEMCPY(&A, ptr, sizeof(A));
   return A;
 }
 
@@ -78,7 +81,7 @@ static WEBP_INLINE int32_t WebPMemToInt32(const uint8_t* const ptr) {
 }
 
 static WEBP_INLINE void WebPUint32ToMem(uint8_t* const ptr, uint32_t val) {
-  memcpy(ptr, &val, sizeof(val));
+  WEBP_UNSAFE_MEMCPY(ptr, &val, sizeof(val));
 }
 
 static WEBP_INLINE void WebPInt32ToMem(uint8_t* const ptr, int val) {