2026-10-18 20:01:18 网络安全文章来源：ZONE.CI 全球网 0 阅读模式

cover_image

Claude Code 隐蔽遥测行为分析

原创

吾爱pojie
吾爱pojie

吾爱破解论坛

2026年7月2日 15:04
北京

在小说阅读器读本章

去阅读

作者论坛账号：andy\_wang425

前言

今天刷到一篇文章说 Claude Code 会检测用户是否使用代理和所在时区，然后修改 system prompt 中的几个字符偷偷上报这些信息。原帖在 Reddit，标题是“Anthropic embedded spyware in Claude Code — and attempted to hide it from you”。以下是我的验证过程。我的环境是 Windows 11，Claude Code 版本为 2.1.196，通过 npm 安装。

过程

提取 JS 代码

Claude Code 是通过 Bun 打包为二进制的，因此首先要做的是把二进制还原为 javascript 代码。我使用的工具来源于 Github Gist（TheCjw/extract\_bun.js），代码如下：

extract\_bun.js

 复制代码 隐藏代码
#!/usr/bin/env bun

import { mkdirSync, readFileSync, writeFileSync } from\"node:fs\";
import path from\"node:path\";

constTRAILER = Buffer.from(\"\
---- Bun! ----\
\");
constBUN_SECTION_NAME = \".bun\";
constOFFSET_STRUCT_SIZE = 32;
constMODULE_RECORD_SIZE = 52;

constENCODINGS = {
0: \"binary\",
1: \"latin1\",
2: \"utf8\",
};

constMODULE_FORMATS = {
0: \"none\",
1: \"esm\",
2: \"cjs\",
};

constSIDES = {
0: \"server\",
1: \"client\",
};

constLOADERS = {
0: \"jsx\",
1: \"js\",
2: \"ts\",
3: \"tsx\",
4: \"css\",
5: \"file\",
6: \"json\",
7: \"jsonc\",
8: \"toml\",
9: \"wasm\",
10: \"napi\",
11: \"base64\",
12: \"dataurl\",
13: \"text\",
14: \"bunsh\",
15: \"sqlite\",
16: \"sqlite_embedded\",
17: \"html\",
18: \"yaml\",
19: \"json5\",
20: \"md\",
};

functiondie(message) {
thrownewError(`error: ${message}`);
}

functionreadUInt64LE(buffer, offset, what) {
const value = buffer.readBigUInt64LE(offset);
if (value > BigInt(Number.MAX_SAFE_INTEGER)) {
    die(`${what} exceeds JavaScript safe integer range: ${value}`);
  }
returnNumber(value);
}

functioncheckedSlice(buffer, offset, size, what) {
if (offset < 0 || size < 0 || offset + size > buffer.length) {
    die(
      `${what} out of bounds: offset=${offset} size=${size} blob_size=${buffer.length}`,
    );
  }
return buffer.subarray(offset, offset + size);
}

functionreadCString(buffer, offset, size, what) {
const data = checkedSlice(buffer, offset, size, what);
return data.length > 0 && data[data.length - 1] === 0
    ? data.subarray(0, -1)
    : data;
}

functiondecodeUtf8(buffer) {
return buffer.toString(\"utf8\").replace(/\\u0000+$/u, \"\");
}

constELF_MAGIC = Buffer.from([0x7f, 0x45, 0x4c, 0x46]);

// PE field offsets
constPE_OFFSET_PTR = 0x3c; // DOS header: offset of PE signature
constPE_NUM_SECTIONS_OFF = 0x06; // COFF header: NumberOfSections (relative to PE sig)
constPE_OPT_HDR_SIZE_OFF = 0x14; // COFF header: SizeOfOptionalHeader (relative to PE sig)
constPE_COFF_HDR_SIZE = 0x18; // size of COFF header (PE sig + COFF fields)
constPE_OPT_MAGIC_OFF = 0x18; // Optional header Magic (relative to PE sig)
constPE_OPT_MAGIC_PE32P = 0x20b; // PE32+ (64-bit)
constPE_SECTION_ENTRY_SIZE = 0x28; // 40 bytes per section header entry
constPE_SECT_RAW_SIZE_OFF = 0x10; // SizeOfRawData (relative to section entry)
constPE_SECT_RAW_OFF_OFF = 0x14; // PointerToRawData (relative to section entry)
constPE_SECT_NAME_LEN = 0x08; // section name field length

// ELF field offsets
constELF_EI_CLASS = 0x04; // e_ident[EI_CLASS]: 1=32-bit, 2=64-bit
constELF_EI_DATA = 0x05; // e_ident[EI_DATA]:  1=LE, 2=BE
constELF_CLASS_64 = 0x02;
constELF_DATA_LE = 0x01;
// Elf64_Ehdr field offsets
constELF64_E_SHOFF = 0x28; // u64
constELF64_E_SHENTSIZE = 0x3a; // u16
constELF64_E_SHNUM = 0x3c; // u16
constELF64_E_SHSTRNDX = 0x3e; // u16
// Elf32_Ehdr field offsets
constELF32_E_SHOFF = 0x20; // u32
constELF32_E_SHENTSIZE = 0x2e; // u16
constELF32_E_SHNUM = 0x30; // u16
constELF32_E_SHSTRNDX = 0x32; // u16
// Elf64_Shdr field offsets (relative to shdr entry)
constELF64_SH_NAME = 0x00; // u32
constELF64_SH_OFFSET = 0x18; // u64
constELF64_SH_SIZE = 0x20; // u64
// Elf32_Shdr field offsets (relative to shdr entry)
constELF32_SH_NAME = 0x00; // u32
constELF32_SH_OFFSET = 0x10; // u32
constELF32_SH_SIZE = 0x14; // u32

// Mach-O magic values (read as big-endian u32)
constMH_MAGIC = 0xfeedface; // 32-bit LE
constMH_CIGAM = 0xcefaedfe; // 32-bit BE
constMH_MAGIC_64 = 0xfeedfacf; // 64-bit LE
constMH_CIGAM_64 = 0xcffaedfe; // 64-bit BE
constFAT_MAGIC = 0xcafebabe;
constFAT_CIGAM = 0xbebafeca;
constFAT_MAGIC_64 = 0xcafebabf;
constFAT_CIGAM_64 = 0xbfbafeca;
// fat_header field offsets (always BE)
constFAT_NFAT_ARCH_OFF = 0x04; // u32
constFAT_ARCH_ENTRY_SIZE = 0x14; // 20 bytes
constFAT_ARCH_TABLE_OFF = 0x08; // first fat_arch
constFAT_ARCH_OFFSET_OFF = 0x08; // offset field within fat_arch entry
// mach_header field offsets
constMACH_NCMDS_OFF = 0x10; // u32 (relative to thin header base)
constMACH_HDR_SIZE_32 = 0x1c; // 28 bytes
constMACH_HDR_SIZE_64 = 0x20; // 32 bytes
constLC_SEGMENT = 0x01;
constLC_SEGMENT_64 = 0x19;
constLC_CMD_OFF = 0x00; // u32 cmd
constLC_CMDSIZE_OFF = 0x04; // u32 cmdsize
constLC_SEGNAME_OFF = 0x08; // char[16]
constLC_SEGNAME_LEN = 0x10;
// segment_command (32-bit) field offsets
constSEG32_NSECTS_OFF = 0x30; // u32
constSEG32_SECTS_OFF = 0x38; // first section entry
constSECT32_ENTRY_SIZE = 0x44; // 68 bytes
constSECT32_SECTNAME_OFF = 0x00; // char[16]
constSECT32_SIZE_OFF = 0x1c; // u32
constSECT32_OFFSET_OFF = 0x20; // u32
// segment_command_64 field offsets
constSEG64_NSECTS_OFF = 0x40; // u32
constSEG64_SECTS_OFF = 0x48; // first section_64 entry
constSECT64_ENTRY_SIZE = 0x50; // 80 bytes
constSECT64_SECTNAME_OFF = 0x00; // char[16]
constSECT64_SIZE_OFF = 0x28; // u64
constSECT64_OFFSET_OFF = 0x30; // u32

functionfindBunElfSection(buf) {
if (buf.length < 0x40) die(\"ELF file too small\");
const is64 = buf[ELF_EI_CLASS] === ELF_CLASS_64;
if (!is64) die(\"ELF: only 64-bit executables are supported\");
const le = buf[ELF_EI_DATA] === ELF_DATA_LE;
const dv = newDataView(buf.buffer, buf.byteOffset, buf.byteLength);
constr16 = (o) => dv.getUint16(o, le);
constr32 = (o) => dv.getUint32(o, le);
constr64 = (o) => {
    const lo = le ? dv.getUint32(o, true) : dv.getUint32(o + 4, false);
    const hi = le ? dv.getUint32(o + 4, true) : dv.getUint32(o, false);
    return hi * 0x100000000 + lo;
  };

const shoff = r64(ELF64_E_SHOFF);
const shentsize = r16(ELF64_E_SHENTSIZE);
const shnum = r16(ELF64_E_SHNUM);
const shstrndx = r16(ELF64_E_SHSTRNDX);

checkedSlice(buf, shoff, shnum * shentsize, \"ELF section header table\");

const shstrOff = shoff + shstrndx * shentsize;
const shstrFileOff = r64(shstrOff + ELF64_SH_OFFSET);
const shstrSize = r64(shstrOff + ELF64_SH_SIZE);
checkedSlice(buf, shstrFileOff, shstrSize, \"ELF shstrtab\");

const matches = [];
for (let i = 0; i < shnum; i++) {
    const shdrOff = shoff + i * shentsize;
    const nameIdx = r32(shdrOff + ELF64_SH_NAME);
    let nameEnd = shstrFileOff + nameIdx;
    while (nameEnd < buf.length && buf[nameEnd] !== 0) nameEnd++;
    const name = buf.toString(\"ascii\", shstrFileOff + nameIdx, nameEnd);
    if (name === \".bun\") {
      const rawOffset = r64(shdrOff + ELF64_SH_OFFSET);
      const rawSize = r64(shdrOff + ELF64_SH_SIZE);
      checkedSlice(buf, rawOffset, rawSize, \".bun section raw data\");
      matches.push({
        format: \"ELF\",
        name,
        rawOffset,
        rawSize,
        data: buf.subarray(rawOffset, rawOffset + rawSize),
      });
    }
  }

if (matches.length === 0) die(\"ELF has no .bun section\");
if (matches.length > 1) die(\"ELF has multiple .bun sections\");
return matches[0];
}

functionparseThinMacho(buf, base) {
if (buf.length < base + MACH_HDR_SIZE_32) die(\"Mach-O thin header too small\");
const magic = buf.readUInt32LE(base);
const be = magic === MH_CIGAM || magic === MH_CIGAM_64;
const is64 = magic === MH_MAGIC_64 || magic === MH_CIGAM_64;
if (!is64) die(\"Mach-O: only 64-bit executables are supported\");
constr32 = (o) => (be ? buf.readUInt32BE(o) : buf.readUInt32LE(o));

const ncmds = r32(base + MACH_NCMDS_OFF);
const hdrSize = MACH_HDR_SIZE_64;
const matches = [];

let off = base + hdrSize;
for (let i = 0; i < ncmds; i++) {
    const cmd = r32(off + LC_CMD_OFF);
    const cmdsize = r32(off + LC_CMDSIZE_OFF);
    if (cmdsize < 8) die(\"Mach-O load command size too small\");
    if (cmd === LC_SEGMENT_64) {
      const segname = buf
        .toString(
          \"ascii\",
          off + LC_SEGNAME_OFF,
          off + LC_SEGNAME_OFF + LC_SEGNAME_LEN,
        )
        .replace(/\\0+$/, \"\");
      if (segname === \"__BUN\") {
        const nsects = r32(off + SEG64_NSECTS_OFF);
        for (let j = 0; j < nsects; j++) {
          const s = off + SEG64_SECTS_OFF + j * SECT64_ENTRY_SIZE;
          const sectname = buf
            .toString(
              \"ascii\",
              s + SECT64_SECTNAME_OFF,
              s + SECT64_SECTNAME_OFF + LC_SEGNAME_LEN,
            )
            .replace(/\\0+$/, \"\");
          if (sectname === \"__bun\") {
            const rawSize = Number(buf.readBigUInt64LE(s + SECT64_SIZE_OFF));
            const rawOffset = be
              ? buf.readUInt32BE(s + SECT64_OFFSET_OFF)
              : buf.readUInt32LE(s + SECT64_OFFSET_OFF);
            matches.push({ rawOffset, rawSize });
          }
        }
      }
    }
    off += cmdsize;
  }
return matches;
}

functionfindBunMachoSection(buf) {
const magic = buf.readUInt32BE(0);
constFAT_SET = newSet([FAT_MAGIC, FAT_CIGAM, FAT_MAGIC_64, FAT_CIGAM_64]);
const allMatches = [];

if (FAT_SET.has(magic)) {
    const nfat = buf.readUInt32BE(FAT_NFAT_ARCH_OFF);
    for (let i = 0; i < nfat; i++) {
      const archEntry = FAT_ARCH_TABLE_OFF + i * FAT_ARCH_ENTRY_SIZE;
      const archFileOff = buf.readUInt32BE(archEntry + FAT_ARCH_OFFSET_OFF);
      allMatches.push(...parseThinMacho(buf, archFileOff));
    }
  } else {
    allMatches.push(...parseThinMacho(buf, 0));
  }

if (allMatches.length === 0) die(\"Mach-O has no __BUN,__bun section\");
if (allMatches.length > 1) die(\"Mach-O has multiple __BUN,__bun sections\");
const { rawOffset, rawSize } = allMatches[0];
checkedSlice(buf, rawOffset, rawSize, \"__BUN,__bun section raw data\");
return {
    format: \"Mach-O\",
    name: \"__BUN,__bun\",
    rawOffset,
    rawSize,
    data: buf.subarray(rawOffset, rawOffset + rawSize),
  };
}

functionfindBunPeSection(buf) {
if (buf.length < 0x40 || buf.toString(\"ascii\", 0, 2) !== \"MZ\") {
    die(\"invalid PE file: missing MZ header\");
  }

const peOffset = buf.readUInt32LE(PE_OFFSET_PTR);
checkedSlice(buf, peOffset, PE_COFF_HDR_SIZE, \"PE header\");
if (buf.toString(\"ascii\", peOffset, peOffset + 4) !== \"PE\\0\\0\") {
    die(\"invalid PE file: missing PE signature\");
  }

const optMagic = buf.readUInt16LE(peOffset + PE_OPT_MAGIC_OFF);
if (optMagic !== PE_OPT_MAGIC_PE32P) {
    die(
      `PE: only 64-bit (PE32+) executables are supported (optional header magic: 0x${optMagic.toString(16)})`,
    );
  }

const numberOfSections = buf.readUInt16LE(peOffset + PE_NUM_SECTIONS_OFF);
const sizeOfOptHdr = buf.readUInt16LE(peOffset + PE_OPT_HDR_SIZE_OFF);
const sectionTableOffset = peOffset + PE_COFF_HDR_SIZE + sizeOfOptHdr;
checkedSlice(
    buf,
    sectionTableOffset,
    numberOfSections * PE_SECTION_ENTRY_SIZE,
    \"section table\",
  );

const matches = [];
for (let index = 0; index < numberOfSections; index++) {
    const sectionOffset = sectionTableOffset + index * PE_SECTION_ENTRY_SIZE;
    const rawName = buf.subarray(
      sectionOffset,
      sectionOffset + PE_SECT_NAME_LEN,
    );
    const nulIndex = rawName.indexOf(0);
    const name = rawName
      .subarray(0, nulIndex === -1 ? rawName.length : nulIndex)
      .toString(\"ascii\");
    const rawSize = buf.readUInt32LE(sectionOffset + PE_SECT_RAW_SIZE_OFF);
    const rawOffset = buf.readUInt32LE(sectionOffset + PE_SECT_RAW_OFF_OFF);

    if (name === BUN_SECTION_NAME) {
      checkedSlice(buf, rawOffset, rawSize, \".bun section raw data\");
      matches.push({
        format: \"PE\",
        name,
        rawOffset,
        rawSize,
        data: buf.subarray(rawOffset, rawOffset + rawSize),
      });
    }
  }

if (matches.length === 0) die(\"PE has no .bun section\");
if (matches.length > 1) die(\"PE has multiple .bun sections\");
return matches[0];
}

constMACHO_MAGIC_SET = newSet([
FAT_MAGIC,
FAT_CIGAM,
FAT_MAGIC_64,
FAT_CIGAM_64,
MH_MAGIC,
MH_CIGAM,
MH_MAGIC_64,
MH_CIGAM_64,
]);

functionfindBunSection(buf) {
if (buf.length < 4) die(\"file too small to determine format\");
const magic = buf.readUInt32BE(0);
if (buf.subarray(0, 4).equals(ELF_MAGIC)) returnfindBunElfSection(buf);
if (MACHO_MAGIC_SET.has(magic)) returnfindBunMachoSection(buf);
returnfindBunPeSection(buf);
}

functionparsePayload(sectionData) {
if (sectionData.length < 8) {
    die(\".bun section is too small for length prefix\");
  }
const payloadSize = readUInt64LE(sectionData, 0, \".bun payload length\");
if (payloadSize + 8 > sectionData.length) {
    die(
      `.bun payload length exceeds raw section: payload=${payloadSize} raw=${sectionData.length}`,
    );
  }

const payload = sectionData.subarray(8, 8 + payloadSize);
if (payload.length < OFFSET_STRUCT_SIZE + TRAILER.length) {
    die(\".bun payload is too small for Offsets + trailer\");
  }
if (!payload.subarray(payload.length - TRAILER.length).equals(TRAILER)) {
    die(\".bun payload trailer mismatch\");
  }
return { payloadSize, payload };
}

functionparseOffsets(payload) {
const start = payload.length - TRAILER.length - OFFSET_STRUCT_SIZE;
const offsets = {
    byte_count: readUInt64LE(payload, start, \"byte_count\"),
    modules_offset: payload.readUInt32LE(start + 8),
    modules_size: payload.readUInt32LE(start + 12),
    entry_point_id: payload.readUInt32LE(start + 16),
    compile_exec_argv_offset: payload.readUInt32LE(start + 20),
    compile_exec_argv_size: payload.readUInt32LE(start + 24),
    flags: payload.readUInt32LE(start + 28),
  };

if (offsets.byte_count > payload.length) {
    die(
      `byte_count exceeds payload: ${offsets.byte_count} > ${payload.length}`,
    );
  }
if (offsets.modules_size % MODULE_RECORD_SIZE !== 0) {
    die(
      `modules table size is not a multiple of ${MODULE_RECORD_SIZE}: ${offsets.modules_size}`,
    );
  }
checkedSlice(
    payload,
    offsets.modules_offset,
    offsets.modules_size,
    \"modules table\",
  );
checkedSlice(
    payload,
    offsets.compile_exec_argv_offset,
    offsets.compile_exec_argv_size,
    \"compile_exec_argv\",
  );
return offsets;
}

functionreadPointerPairs(record) {
const pairs = [];
for (let offset = 0; offset < 48; offset += 8) {
    pairs.push([record.readUInt32LE(offset), record.readUInt32LE(offset + 4)]);
  }
return pairs;
}

functionparseFiles(payload, offsets, sectionFileOffset) {
const files = [];
const count = offsets.modules_size / MODULE_RECORD_SIZE;
if (offsets.entry_point_id >= count) {
    die(`entry_point_id out of range: ${offsets.entry_point_id} >= ${count}`);
  }

const table = checkedSlice(
    payload,
    offsets.modules_offset,
    offsets.modules_size,
    \"modules table\",
  );
for (let index = 0; index < count; index += 1) {
    const record = table.subarray(
      index * MODULE_RECORD_SIZE,
      (index + 1) * MODULE_RECORD_SIZE,
    );
    const pointers = readPointerPairs(record);
    const encodingId = record.readUInt8(48);
    const loaderId = record.readUInt8(49);
    const moduleFormatId = record.readUInt8(50);
    const sideId = record.readUInt8(51);

    const [nameOffset, nameSize] = pointers[0];
    const [contentOffset, contentSize] = pointers[1];
    const [sourcemapOffset, sourcemapSize] = pointers[2];
    const [bytecodeOffset, bytecodeSize] = pointers[3];
    const [moduleInfoOffset, moduleInfoSize] = pointers[4];
    const [bytecodeOriginPathOffset, bytecodeOriginPathSize] = pointers[5];

    const name = decodeUtf8(
      readCString(payload, nameOffset, nameSize, `module[${index}].name`),
    );
    checkedSlice(
      payload,
      contentOffset,
      contentSize,
      `module[${index}].contents`,
    );
    if (sourcemapSize) {
      checkedSlice(
        payload,
        sourcemapOffset,
        sourcemapSize,
        `module[${index}].sourcemap`,
      );
    }
    if (bytecodeSize) {
      checkedSlice(
        payload,
        bytecodeOffset,
        bytecodeSize,
        `module[${index}].bytecode`,
      );
    }
    if (moduleInfoSize) {
      checkedSlice(
        payload,
        moduleInfoOffset,
        moduleInfoSize,
        `module[${index}].module_info`,
      );
    }

    let bytecodeOriginPath = \"\";
    if (bytecodeOriginPathSize) {
      bytecodeOriginPath = decodeUtf8(
        readCString(
          payload,
          bytecodeOriginPathOffset,
          bytecodeOriginPathSize,
          `module[${index}].bytecode_origin_path`,
        ),
      );
    }

    files.push({
      index,
      entry: index === offsets.entry_point_id,
      name,
      content_payload_offset: contentOffset,
      content_file_offset: sectionFileOffset + 8 + contentOffset,
      content_size: contentSize,
      loader: LOADERS[loaderId] ?? `unknown(${loaderId})`,
      encoding: ENCODINGS[encodingId] ?? `unknown(${encodingId})`,
      module_format:
        MODULE_FORMATS[moduleFormatId] ?? `unknown(${moduleFormatId})`,
      side: SIDES[sideId] ?? `unknown(${sideId})`,
      bytecode_payload_offset: bytecodeSize ? bytecodeOffset : null,
      bytecode_size: bytecodeSize,
      module_info_payload_offset: moduleInfoSize ? moduleInfoOffset : null,
      module_info_size: moduleInfoSize,
      bytecode_origin_path: bytecodeOriginPath,
      sourcemap_payload_offset: sourcemapSize ? sourcemapOffset : null,
      sourcemap_size: sourcemapSize,
    });
  }
return files;
}

functioniterLimited(files, limit) {
return limit == null ? files : files.slice(0, limit);
}

constBUN_VIRTUAL_ROOTS = newSet([\"~BUN\", \"$bunfs\"]);

functionoutputRelativePath(name) {
const normalized = name.replaceAll(\"\\\\\", \"/\");
let parts = normalized
    .split(\"/\")
    .filter((p) => p !== \"\" && p !== \".\" && p !== \"..\" && !p.endsWith(\":\"));
if (parts.length > 0 && BUN_VIRTUAL_ROOTS.has(parts[0])) {
    parts = parts.slice(1);
    if (parts.length > 0 && parts[0] === \"root\") parts = parts.slice(1);
  }
return parts.length === 0 ? \"unnamed\" : path.join(...parts);
}

functionextractFiles(payload, files, outputDir) {
const extracted = [];
mkdirSync(outputDir, { recursive: true });
const resolvedOut = path.resolve(outputDir);

for (const file of files) {
    const relativePath = outputRelativePath(file.name);
    const destination = path.join(outputDir, relativePath);
    const resolved = path.resolve(destination);
    if (
      resolved !== resolvedOut &&
      !resolved.startsWith(resolvedOut + path.sep)
    ) {
      die(
        `refusing to write outside output dir: ${JSON.stringify(file.name)} -> ${destination}`,
      );
    }
    mkdirSync(path.dirname(destination), { recursive: true });
    const content = checkedSlice(
      payload,
      file.content_payload_offset,
      file.content_size,
      `module[${file.index}].contents`,
    );
    writeFileSync(destination, content);
    extracted.push({
      index: file.index,
      entry: file.entry,
      original_name: file.name,
      saved_path: destination,
      content_payload_offset: file.content_payload_offset,
      content_file_offset: file.content_file_offset,
      content_size: file.content_size,
      loader: file.loader,
      side: file.side,
      bytecode_size: file.bytecode_size,
      module_info_size: file.module_info_size,
      bytecode_origin_path: file.bytecode_origin_path,
    });
  }

const manifestPath = path.join(outputDir, \"manifest.json\");
writeFileSync(
    manifestPath,
    `${JSON.stringify(extracted, null, 2)}\
`,
    \"utf8\",
  );
return extracted;
}

functionprintText(exePath, section, payloadSize, offsets, files, limit) {
console.log(`exe: ${exePath}`);
console.log(
    `${section.format}${section.name}: raw_offset=0x${section.rawOffset.toString(16)} raw_size=${section.rawSize} payload_size=${payloadSize}`,
  );
console.log(
    `graph: byte_count=${offsets.byte_count} modules_offset=${offsets.modules_offset} modules_size=${offsets.modules_size} module_count=${files.length} entry_point_id=${offsets.entry_point_id} flags=0x${offsets.flags.toString(16)}`,
  );
console.log(
    \"index\\tentry\\tloader\\tside\\tcontent_payload_offset\\tcontent_file_offset\\tcontent_size\\tname\",
  );
for (const file ofiterLimited(files, limit)) {
    console.log(
      `${file.index}\\t${Number(file.entry)}\\t${file.loader}\\t${file.side}\\t${file.content_payload_offset}\\t0x${file.content_file_offset.toString(16)}\\t${file.content_size}\\t${file.name}`,
    );
  }
if (limit != null && files.length > limit) {
    console.log(
      `... ${files.length - limit} more file(s) omitted; use --limit 0 for all`,
    );
  }
}

functionparseLimit(value) {
const n = Number.parseInt(value, 10);
if (!Number.isInteger(n) || n < 0)
    die(\"--limit must be a non-negative integer\");
return n === 0 ? null : n;
}

functionparseSubArgs(argv, validFlags) {
// Returns { exe, flags... }; exe is required positional arg
const result = { exe: null };
for (let i = 0; i < argv.length; i++) {
    const arg = argv[i];
    if (arg === \"-h\" || arg === \"--help\") {
      result._help = true;
      return result;
    }
    const flag = validFlags.find(
      (f) => arg === f.flag || (f.short && arg === f.short),
    );
    if (flag) {
      if (flag.takesValue) {
        i++;
        if (i >= argv.length) die(`${arg} requires a value`);
        result[flag.key] = flag.parse ? flag.parse(argv[i]) : argv[i];
      } else {
        result[flag.key] = true;
      }
    } elseif (arg.startsWith(\"-\")) {
      die(`unknown option: ${arg}`);
    } elseif (result.exe === null) {
      result.exe = arg;
    } else {
      die(`unexpected positional argument: ${arg}`);
    }
  }
return result;
}

functionprintHelp() {
console.log(`Usage: bun extract_bun.js <command> EXE [options]

Parse Bun build --compile embedded files from PE/ELF/Mach-O executables.

Commands:
  print EXE    Print embedded module graph information
  extract EXE  Extract embedded files and write manifest.json

Options for print:
  --limit N    Maximum records to print; 0 means no limit (default: all)
  --json       Emit JSON instead of TSV-like text
  -h, --help   Show this help message

Options for extract:
  -o, --output DIR  Output directory (required)
  -h, --help        Show this help message`);
}

functionprintExtractedText(
  exePath,
  section,
  payloadSize,
  offsets,
  files,
  extracted,
  outputDir,
) {
console.log(`exe: ${exePath}`);
console.log(
    `${section.format}${section.name}: raw_offset=0x${section.rawOffset.toString(16)} raw_size=${section.rawSize} payload_size=${payloadSize}`,
  );
console.log(
    `graph: module_count=${files.length} entry_point_id=${offsets.entry_point_id} flags=0x${offsets.flags.toString(16)}`,
  );
console.log(\"index\\tentry\\tloader\\tside\\tsize\\tsaved_path\\toriginal_name\");
for (const file of extracted) {
    console.log(
      `${file.index}\\t${Number(file.entry)}\\t${file.loader}\\t${file.side}\\t${file.content_size}\\t${file.saved_path}\\t${file.original_name}`,
    );
  }
console.log(`extracted: ${extracted.length} file(s) -> ${outputDir}`);
console.log(`manifest: ${path.join(outputDir, \"manifest.json\")}`);
}

functionmain(argv = process.argv.slice(2)) {
const [subcmd, ...rest] = argv;

if (!subcmd || subcmd === \"-h\" || subcmd === \"--help\") {
    printHelp();
    process.exit(0);
  }

if (subcmd === \"print\") {
    const args = parseSubArgs(rest, [
      { flag: \"--limit\", key: \"limit\", takesValue: true, parse: parseLimit },
      { flag: \"--json\", key: \"json\" },
    ]);
    if (args._help) {
      console.log(
        `Usage: bun extract_bun.js print EXE [--limit N] [--json] [-h]`,
      );
      process.exit(0);
    }
    if (!args.exe) die(\"print requires an EXE argument\");
    const buf = readFileSync(args.exe);
    const section = findBunSection(buf);
    const { payloadSize, payload } = parsePayload(section.data);
    const offsets = parseOffsets(payload);
    const files = parseFiles(payload, offsets, section.rawOffset);
    const limit = args.limit ?? null;
    if (args.json) {
      console.log(
        JSON.stringify(
          {
            exe: args.exe,
            bun_section: {
              format: section.format,
              name: section.name,
              raw_offset: section.rawOffset,
              raw_size: section.rawSize,
              payload_size: payloadSize,
            },
            graph: { ...offsets, module_count: files.length },
            files: iterLimited(files, limit),
            omitted: limit == null ? 0 : Math.max(0, files.length - limit),
            extracted: [],
          },
          null,
          2,
        ),
      );
    } else {
      printText(args.exe, section, payloadSize, offsets, files, limit);
    }
  } elseif (subcmd === \"extract\") {
    const args = parseSubArgs(rest, [
      { flag: \"--output\", short: \"-o\", key: \"output\", takesValue: true },
    ]);
    if (args._help) {
      console.log(`Usage: bun extract_bun.js extract EXE -o DIR [-h]`);
      process.exit(0);
    }
    if (!args.exe) die(\"extract requires an EXE argument\");
    if (!args.output) die(\"extract requires -o/--output DIR\");
    const buf = readFileSync(args.exe);
    const section = findBunSection(buf);
    const { payloadSize, payload } = parsePayload(section.data);
    const offsets = parseOffsets(payload);
    const files = parseFiles(payload, offsets, section.rawOffset);
    const extracted = extractFiles(payload, files, args.output);
    printExtractedText(
      args.exe,
      section,
      payloadSize,
      offsets,
      files,
      extracted,
      args.output,
    );
  } else {
    die(
      `unknown command: ${subcmd}\
Run 'bun extract_bun.js --help' for usage.`,
    );
  }
}

try {
main();
} catch (error) {
console.error(error.message);
  process.exit(1);
}

虽然作者说要用 Bun 运行，但实测用 Node 也不影响。运行以下命令即可把 claude.exe 解包为 js 代码：

 复制代码 隐藏代码
bun extract_bun.js extract claude.exe -o extract

入口模块位于 src/entrypoints/cli.js。可以先把这个文件格式化一下方便后续查看。

关键代码分析

代理检测

 复制代码 隐藏代码
// cli.js:114617
functionArt() {
let e = process.env.ANTHROPIC_BASE_URL;
if (!e) return !0; // 未设代理 → 视为\"第一方\"
returnTrt(e); // 代理主机为 api.anthropic.com 时返回 true
}
// cli.js:114622
functionTrt(e) {
try {
    return [\"api.anthropic.com\"].includes(newURL(e).host);
  } catch {
    return !1;
  }
}

Art() 在 ANTHROPIC_BASE_URL 未设置或指向 api.anthropic.com 时返回 true。这是后续检测的前置判断，只有当该函数返回 false 时才会做后续的域名和时区检测。

核心检测逻辑

 复制代码 隐藏代码
// cli.js:274138
functionWup() {
if (Art()) returnnull; // 直连/官方端点 → 不做任何检测
let e = jup(), // 取 ANTHROPIC_BASE_URL 的 hostname
    t = tkt(), // 取系统时区
    n = t === \"Asia/Shanghai\" || t === \"Asia/Urumqi\";
if (!e) return { known: !1, labKw: !1, cnTZ: n, host: null };
return {
    known: Fup().some((r) => e === r || e.endsWith(\".\" + r)), // 命中域名白名单
    labKw: Bup().some((r) => e.includes(r)), // 命中 AI 实验室关键词
    cnTZ: n,
    host: e,
  };
}

// cli.js:274129  —— jup() 的实现
functionjup() {
let e = process.env.ANTHROPIC_BASE_URL;
if (!e) returnnull;
try {
    returnnewURL(e).hostname.toLowerCase();
  } catch {
    returnnull;
  }
}

// cli.js:12714  —— tkt() 的实现
functiontkt() {
if (!fTr) fTr = Intl.DateTimeFormat().resolvedOptions().timeZone;
return fTr;
}

核心检测逻辑在 Wup()。如果发现用户没走官方端点，就会进一步检测域名和时区，最后返回包含 4 个字段的检测结果：

known

：ANTHROPIC_BASE_URL 的域名是否命中域名白名单
* labKw

：ANTHROPIC_BASE_URL 的域名是否命中 AI 实验室关键词
* cnTZ

：当前系统时区是不是中国的（判断时区是否为上海或乌鲁木齐）
* host

：ANTHROPIC_BASE_URL 的域名

撇号与日期格式隐写

 复制代码 隐藏代码
// cli.js:274151
functionVup(e, t) {
// e = known(域名命中)  t = labKw(ai实验室命中)
if (!e && !t) return\"'\"; // 都不命中 → 普通 ASCII 撇号 U+0027
if (e && !t) return\"\\u2019\"; // 仅域名   → ’ RIGHT SINGLE QUOTATION MARK
if (!e && t) return\"\\u02BC\"; // 仅实验室 → ʼ MODIFIER LETTER APOSTROPHE
return\"\\u02B9\"; // 都命中   → ʹ MODIFIER LETTER PRIME
}

// cli.js:274157
functionLla(e) {
// e = 当前日期 \"YYYY-MM-DD\"
let t = Wup(),
    n = Vup(t?.known ?? !1, t?.labKw ?? !1),
    r = t?.cnTZ ? e.replaceAll(\"-\", \"/\") : e; // 中国时区 → YYYY/MM/DD
return`Today${n}s date is ${r}.`;
}

Claude Code 的系统提示词中有一段 Today's date is 当前日期.，而这两个函数会修改其中的撇号和日期格式。

正常情况下，当前日期应该是 YYYY-MM-DD 的格式，但如果之前的检测发现用户是中国时区，就会把 - 替换成 /。此外还会根据是否命中域名白名单（known）和是否命中 AI 实验室关键词（labKw）来选择 Today's 中的撇号，一共有 4 中情况：

known(域名)	labKw(实验室)	输出字符	Unicode
✗	✗	`'`	U+0027
✓	✗	`’`	U+2019
✗	✓	`ʼ`	U+02BC
✓	✓	`ʹ`	U+02B9

这 4 种撇号在视觉上差别很小，因此用户很难发现异常。

XOR/base64 混淆

 复制代码 隐藏代码
// cli.js:274163
varNup = 91,
  $up = \"ODV3KDo1MC46...\", // 域名白名单（base64+XOR91）
Uup = \"Pz4+Kyg+PjB3...\", // AI 实验室关键词
Fup,
Bup;

varOla = E(() => {
ea();
H0();
Ms();
  ((Fup = wn(() =>Mla($up))), // 惰性解码并缓存
    (Bup = wn(() =>Mla(Uup))));
});

// cli.js:274123
functionMla(e) {
let t = Buffer.from(e, \"base64\"),
    n = \"\";
for (let r of t) n += String.fromCharCode(r ^ Nup); // 逐字节 XOR 0x5B(=91)
return n.split(\",\");
}

之前提到的域名白名单和 AI 实验室关键词是被加密保存的。解密方式是先 Base64 解码，然后逐字节异或一个 key 91。我们可以写个简单的脚本来解密：

 复制代码 隐藏代码
constNup = 91;
functionMla(e) {
let t = Buffer.from(e, \"base64\"),
    n = \"\";
for (let r of t) n += String.fromCharCode(r ^ Nup);
return n.split(\",\");
}

const $up =
    \"ODV3KDo1MC46MnU4NDZ3NT4vPjooPnU4NDZ3am1odTg0Nnc5OjI/LnYyNS91ODQ2dzk6Mj8udTg0Nnc6NzI5Ojk6djI1OHU4NDZ3OjcyKzoidTg0Nnc6NS88KTQuK3YyNTh1ODV3MC46MigzNC51ODQ2dzkiLz4/OjU4PnU1Pi93IzI6NDM0NTwoMy51ODQ2dzgvKTIrODQpK3U4NDZ3MT91ODQ2dzE/ODc0Lj91ODQ2dzkyNzI5MjcydTg0dzI9NyIvPjB1ODQ2dygvPis9LjV2MjU4dTg0Nnc6NzIiLjU4KHU4NDZ3ODV2KDM6NTwzOjJ1PTg6Kyt1KS41dzg1djk+MjEyNTx1PTg6Kyt1KS41dyM6NjI1MjZ1ODQ2dzY0NDUoMzQvdToydzo1Iik0Li8+KXUvNCt3Kzo4MCI6KzJ1ODQ2dzoyODQ/PjYyKSk0KXU4NDZ3OjI8NDg0Pz51ODQ2dzM0NTwoMzo1dTg0NncyLDM6Nz44NzQuP3U4NDZ3PzM4ND8+KXU1Pi93Nz42NDU8Ky91LzQrdyEzMjMuMjorMnUvNCt3MjUvKDI8dTU+L3czMjwzdj0yLT52OjJ1IyIhdzg3NC4/KCw6InU1Pi93byg6KzJ1ODQ2d25pYmJtanU4NDZ3Y2NiYm11ODc0Lj93Y2M4ND8+dToyd2NjODQ/PnU0KTx3Ymo4ND8+dSspNHdiYmlpaG11IyIhdzoydTg0Pz4qOip1ODQ2dzoydTMiOTwhKHU4NDZ3OjJ1MDEtMzN1ODQ2dzoyODo1OisydTg0Nnc6Mjg0PzI1PHUoM3c6Mj06KC91KDIvPnc6MjMuOTYyI3U4NDZ3OjU2NCkidTg0Nnc6KzJ1bmlraWtoa3UjIiF3OisydTo5NzoydS80K3c6KzJ1OTI6NSMyPnU6Mnc6KzJ1OTcvOCJ1OjJ3OisydTgrOigodTg4dzorMnU/Pi1jY3UvPjgzdzorMnU/KT46Njw+KXU4NDZ3OisydT4jKzo1KDI0NXU4MzovdzorMnU8Lj46MnU4NDZ3OisydTM0Nz86MnUvNCt3OisydTIwLjU4ND8+dTg4dzorMnU3ODQ1OjJ1ODQ2dzorMnU3MjUwOisydTQpPHc6KzJ1NjA+OjJ1ODQ2dzorMnU1PjA0OisydTg0Nnc6KzJ1NDoyKyk0dTg0Nnc6KzJ1KS4iLjV1PS41dzorMnUoKDQrPjV1LzQrdzorMnUvLnYhMnU4NDZ3OisydS48NyI4Oi91ODh3OisydS1odTg2dzorMnUsMzovOjJ1ODh3OisydSwrPCEodS80K3c6KzJ1Iy8idTorK3c6KzJ1Ii4+PDc+dTg0Nnc6KzJ1ISEiLnU2Pnc6KzI2OikvdToydzorMispNHU2OiI1NClqa2lvdTcyLT53OisyIjJ1ODQ2dzorKzciMXUzMjorMnUvNCt3Oi48Ni41L3U4NDZ3OW8udSohIXUyNHc4NzouPz8idTg0Nnc4NzouPz52ODQ/PnYzLjl1Oisrdzg3Oi4/PnY0Ky4odS80K3c4NzouPz4yPz51NT4vdzg0dSI+KHUtPHc4ND8+dSw+NSw+NXY6MnU4NDZ3ODQ/PnUjdjoyNHU4NDZ3ODQ/PjI3Ojl1ODQ2dzguOT41OD51ODQ2dz8+PispNC4vPil1LzQrdz8yNjopOiJ1ODQ2dz82IzorMnU4NDZ3PzQ4KHU6Mjw4aT91ODQ2dz8uODA4ND8yNTx1ODQ2dz0wdTMoMywwdTQpPHc9NzorODQ/PnU4NDZ3PTQjODQ/PnUzKDMsMHU0KTx3PTQjODQ/PnUpMTF1ODh3PS43MnUzIzJ1Nj53PD4vPDQ6KzJ1ODQ2dzwrL3UhMzIhPjU8IT41PHU4NDZ3PCsvPDQ/dTg3NC4/dzwrLzA+InU+LnU0KTx3PCsvKzoidSgvNCk+dzM/PCg5dTg0NnczPjU6KzJ1LzQrdzI1KC84NCsyNzQvdjorMnU4NDZ3MT41MiI6dS80K3cxMj4wNC51OjJ3MDx2OisydTg3NC4/dzVqNXU6Mnc1Pix2OisydS5vLSl1ODQ2dzU+LHUjIjgzOi86MnU4NDZ3NDU+djorMnU5Ny84InUvNCt3NDU+dTQ4NDQ3OjJ1ODQ2dzQ1PjorMnUrOjI1Lzk0L3UvNCt3NCs+NXUjMjo0MTI1PDoydTg0Nnc0Kz41ODc6Lj8+dTY+dzQrLih1PCsvLi51ODQ2dys0NzQ6MnUvNCt3KzQ3NDorMnUvNCt3KykyLTU0Pz51ODQ2dyspNCMiOjJ1ODQ2dyoyNSEzMjoydTg0NncpMjwzL3U4ND8+KHcpLjU6NSIvMjY+dTMjMnU2PncoKCg6Mjg0Pz51ODQ2dygvNCk+dSEhIi4odS80K3cvMjo1LzI6NToydSspNHcuMi4yOisydTg0NncuNTI6KzJ1OjJ3LTIrdS41PyIyNTw6KzJ1ODQ2dyw0Nz06MnUvNCt3LCEsdT8+bnU1Pi93LCEsdSsrdS46dyM6Mik0Li8+KXU4NDZ3IzoyIzorMnU4NDZ3IzI6NDMuOisydSgyLz53IzI6NDMuNjI1MnUoMi8+dyMidSs0NzQ6KzJ1ODQ2dyI6NSg/bW1tdTg0NnciOjUoP21tbXUvNCt3Ii41LC51OjJ3Ii41LC51IT46OS4pdTorK3chPjU2LiN1OjI=\",
Uup =
    \"Pz4+Kyg+PjB3NjQ0NSgzNC93NjI1MjY6I3cjOjYyNTI2dyEzMisudzkyPDY0Pz43dzk6MjgzLjo1dygvPis9LjV3a2o6Mnc/OigzKDg0Kz53LTQ3OD4o\";

console.log(\"=== Fup (domain whitelist) ===\");
console.log(JSON.stringify(Mla($up), null, 2));

console.log(\"\
=== Bup (AI lab keywords) ===\");
console.log(JSON.stringify(Mla(Uup), null, 2));

运行结果：

 复制代码 隐藏代码
=== Fup (domain whitelist) ===
[
\"cn\",
\"sankuai.com\",
\"netease.com\",
\"163.com\",
\"baidu-int.com\",
\"baidu.com\",
\"alibaba-inc.com\",
\"alipay.com\",
\"antgroup-inc.cn\",
\"kuaishou.com\",
\"bytedance.net\",
\"xiaohongshu.com\",
\"ctripcorp.com\",
\"jd.com\",
\"jdcloud.com\",
\"bilibili.co\",
\"iflytek.com\",
\"stepfun-inc.com\",
\"aliyuncs.com\",
\"cn-shanghai.fcapp.run\",
\"cn-beijing.fcapp.run\",
\"xaminim.com\",
\"moonshot.ai\",
\"anyrouter.top\",
\"packyapi.com\",
\"aicodemirror.com\",
\"aigocode.com\",
\"hongshan.com\",
\"iwhalecloud.com\",
\"dhcoder.net\",
\"lemongpt.top\",
\"zhihuiapi.top\",
\"intsig.net\",
\"high-five-ai.xyz\",
\"cloudsway.net\",
\"4sapi.com\",
\"529961.com\",
\"88996.cloud\",
\"88code.ai\",
\"88code.org\",
\"91code.pro\",
\"992236.xyz\",
\"ai.codeqaq.com\",
\"ai.hybgzs.com\",
\"ai.kjvhh.com\",
\"aicanapi.com\",
\"aicoding.sh\",
\"aifast.site\",
\"aihubmix.com\",
\"anmory.com\",
\"api.5202030.xyz\",
\"api.ablai.top\",
\"api.bianxie.ai\",
\"api.bltcy.ai\",
\"api.cpass.cc\",
\"api.dev88.tech\",
\"api.dreamger.com\",
\"api.expansion.chat\",
\"api.gueai.com\",
\"api.holdai.top\",
\"api.ikuncode.cc\",
\"api.lconai.com\",
\"api.linkapi.org\",
\"api.mkeai.com\",
\"api.nekoapi.com\",
\"api.oaipro.com\",
\"api.ruyun.fun\",
\"api.ssopen.top\",
\"api.tu-zi.com\",
\"api.uglycat.cc\",
\"api.v3.cm\",
\"api.whatai.cc\",
\"api.wpgzs.top\",
\"api.xty.app\",
\"api.yuegle.com\",
\"api.zzyu.me\",
\"apimart.ai\",
\"apipro.maynor1024.live\",
\"apiyi.com\",
\"applyj.hiapi.top\",
\"augmunt.com\",
\"b4u.qzz.io\",
\"clauddy.com\",
\"claude-code-hub.app\",
\"claude-opus.top\",
\"claudeide.net\",
\"co.yes.vg\",
\"code.wenwen-ai.com\",
\"code.x-aio.com\",
\"codeilab.com\",
\"cubence.com\",
\"deeprouter.top\",
\"dimaray.com\",
\"dmxapi.com\",
\"docs.aigc2d.com\",
\"duckcoding.com\",
\"fk.hshwk.org\",
\"flapcode.com\",
\"foxcode.hshwk.org\",
\"foxcode.rjj.cc\",
\"fuli.hxi.me\",
\"getgoapi.com\",
\"gpt.zhizengzeng.com\",
\"gptgod.cloud\",
\"gptkey.eu.org\",
\"gptpay.store\",
\"hdgsb.com\",
\"henapi.top\",
\"instcopilot-api.com\",
\"jeniya.top\",
\"jiekou.ai\",
\"kg-api.cloud\",
\"n1n.ai\",
\"new-api.u4vr.com\",
\"new.xychatai.com\",
\"one-api.bltcy.top\",
\"one.ocoolai.com\",
\"oneapi.paintbot.top\",
\"open.xiaojingai.com\",
\"openclaude.me\",
\"opus.gptuu.com\",
\"poloai.top\",
\"poloapi.top\",
\"privnode.com\",
\"proxyai.com\",
\"qinzhiai.com\",
\"right.codes\",
\"runanytime.hxi.me\",
\"sssaicode.com\",
\"store.zzyus.top\",
\"tiantianai.pro\",
\"uiuiapi.com\",
\"uniapi.ai\",
\"vip.undyingapi.com\",
\"wolfai.top\",
\"wzw.de5.net\",
\"wzw.pp.ua\",
\"xairouter.com\",
\"xaixapi.com\",
\"xiaohuapi.site\",
\"xiaohumini.site\",
\"xy.poloapi.com\",
\"yansd666.com\",
\"yansd666.top\",
\"yunwu.ai\",
\"yunwu.zeabur.app\",
\"zenmux.ai\"
]

=== Bup (AI lab keywords) ===
[
\"deepseek\",
\"moonshot\",
\"minimax\",
\"xaminim\",
\"zhipu\",
\"bigmodel\",
\"baichuan\",
\"stepfun\",
\"01ai\",
\"dashscope\",
\"volces\"
]

可以看到域名白名单里主要是国内互联网大厂、AI厂商以及API中转站的域名，AI 实验室关键词里则是一些国内头部大模型团队的名字。

注入到系统提示词

Claude Code 最终会把隐写的信息注入到系统提示词：

 复制代码 隐藏代码
// cli.js:279102
currentDate: Lla(GSe()),

Lla() 是刚刚提到的当前日期隐写函数。GSe() 最终指向 Hao()，负责生成 YYYY-MM-DD：

 复制代码 隐藏代码
// cli.js:272925
functionHao() {
  let e = newDate(),
    t = e.getFullYear(),
    n = String(e.getMonth() + 1).padStart(2, \"0\"),
    r = String(e.getDate()).padStart(2, \"0\");
  return`${t}-${n}-${r}`;
}

该 currentDate 字段会被并入会话上下文对象，最终进入每次 API 请求的系统提示词。Anthropic 服务端可凭字符编码（'/’/ʼ/ʹ）与日期分隔符（-//）反推出用户是不是中国的、是否在使用中转站和是不是国内AI厂商的人。

结语

Reddit 原文中的说法均属实。Claude Code 在检测到用户将 ANTHROPIC_BASE_URL 设置为第三方代理时才激活时区和域名检测，且其结果以 Unicode 近形字符隐写的方式藏在系统提示词里，用户和模型都难以察觉。内置的域名白名单和 AI 实验室关键词覆盖大量中转站和国内头部大模型团队，显然是为了防蒸馏和转售。

A÷刚刚在X上也承认了有检测中国用户的机制，还说本来就打算要取消这项实验了。被发现了就说要取消，只能说A÷真的是÷完了。Reddit原帖下面很多人觉得这不算间谍软件（spyware）。这个就见仁见智吧，虽然我觉得说是spyware也没啥问题，但为了客观一点文章标题里还是用了“遥测”这个词。最近 Anthropic 还被爆出在邮件里埋了1x1像素的追踪图片，用户打开邮件时就会泄露ip地址，只能说A÷为了封国人号真是煞费苦心啊。希望国内AI厂商继续加油，在未来某一天干倒A÷。

附件里是我提取并格式化的 cli.js，因为太大了所以只能分卷压缩后上传。解压前需去掉末尾的 .7z。（附件见左下角原文）

*-官方论坛*

www.52pojie.cn

👆👆👆

公众号设置“星标”，您不会错过新的消息通知

如开放注册、精华文章和周边活动等公告