diff --git a/s16code/tools.py b/s16code/tools.py
index f70af95..f302b6c 100644
--- a/s16code/tools.py
+++ b/s16code/tools.py
@@ -6,16 +6,18 @@
 import csv
 import hashlib
 import html
+import ipaddress
 import math
 import operator
 import os
 import re
 import shutil
+import socket
 import sqlite3
 from datetime import date, datetime, timedelta
 from html.parser import HTMLParser
 from pathlib import Path
-from urllib.parse import parse_qs, unquote, urlparse
+from urllib.parse import parse_qs, unquote, urljoin, urlparse
 from zoneinfo import ZoneInfo, ZoneInfoNotFoundError
 
 import httpx
@@ -244,21 +246,64 @@ def date_shift(value: str, days: int) -> dict:
             "weekday": shifted.strftime("%A")}
 
 
-async def fetch_url(url: str, *, max_chars: int = 60_000) -> dict:
+def assert_fetchable_url(url: str) -> None:
+    """Refuse anything that is not a public http(s) target.
+
+    Scheme checks alone are not enough: ``http://127.0.0.1/`` and the cloud
+    metadata link-local address are valid http URLs and would otherwise let a
+    fetched page (or a redirect) read the host the agent is running on.
+    """
     parsed = urlparse(url)
     if parsed.scheme not in {"http", "https"}:
         raise ValueError("fetch_url permits only http(s)")
-    async with httpx.AsyncClient(timeout=30, follow_redirects=True,
+    host = parsed.hostname
+    if not host:
+        raise ValueError("fetch_url URL has no host")
+    try:
+        addresses = [ipaddress.ip_address(host)]
+    except ValueError:
+        try:
+            infos = socket.getaddrinfo(host, parsed.port or (443 if parsed.scheme == "https" else 80),
+                                       type=socket.SOCK_STREAM)
+        except socket.gaierror as error:
+            raise ValueError(f"fetch_url could not resolve host: {host}") from error
+        addresses = [ipaddress.ip_address(info[4][0]) for info in infos]
+        if not addresses:
+            raise ValueError(f"fetch_url could not resolve host: {host}")
+    for address in addresses:
+        if isinstance(address, ipaddress.IPv6Address) and address.ipv4_mapped is not None:
+            address = address.ipv4_mapped
+        if (address.is_loopback or address.is_private or address.is_link_local
+                or address.is_multicast or address.is_reserved or address.is_unspecified):
+            raise ValueError("fetch_url refuses private or loopback addresses")
+
+
+async def fetch_url(url: str, *, max_chars: int = 60_000) -> dict:
+    current = url
+    async with httpx.AsyncClient(timeout=30, follow_redirects=False,
                                  headers={"User-Agent": "GLC-S16/0.3 educational-agent"}) as client:
-        response = await client.get(url)
-        response.raise_for_status()
-    content_type = response.headers.get("content-type", "")
-    if "html" in content_type:
-        parser = _TextExtractor(); parser.feed(response.text); text = parser.text()
-    else:
-        text = response.text
-    return {"url": str(response.url), "status": response.status_code,
-            "content_type": content_type, "text": text[:max_chars], "truncated": len(text) > max_chars}
+        response = None
+        for _ in range(5):
+            assert_fetchable_url(current)
+            response = await client.get(current)
+            if response.is_redirect:
+                location = response.headers.get("location")
+                if not location:
+                    raise ValueError("fetch_url redirect was missing Location")
+                current = urljoin(str(response.url), location)
+                continue
+            response.raise_for_status()
+            break
+        else:
+            raise ValueError("fetch_url followed too many redirects")
+        assert response is not None
+        content_type = response.headers.get("content-type", "")
+        if "html" in content_type:
+            parser = _TextExtractor(); parser.feed(response.text); text = parser.text()
+        else:
+            text = response.text
+        return {"url": str(response.url), "status": response.status_code,
+                "content_type": content_type, "text": text[:max_chars], "truncated": len(text) > max_chars}
 
 
 class _DDGParser(HTMLParser):
diff --git a/tests/test_general_tools.py b/tests/test_general_tools.py
index 8b08223..1412da9 100644
--- a/tests/test_general_tools.py
+++ b/tests/test_general_tools.py
@@ -24,6 +24,25 @@ def test_calculate_supports_useful_arithmetic_without_code_execution():
         calculate("2 ** 100")
 
 
+def test_fetch_url_refuses_private_and_loopback_targets():
+    from s16code.tools import assert_fetchable_url
+
+    assert_fetchable_url("https://1.1.1.1/")
+    with pytest.raises(ValueError, match="http"):
+        assert_fetchable_url("file:///etc/passwd")
+    for hostile in (
+        "http://127.0.0.1/",
+        "http://127.0.0.1:8111/v1/budget",
+        "http://localhost/",
+        "http://10.0.0.1/",
+        "http://192.168.1.1/",
+        "http://169.254.169.254/latest/meta-data/",
+        "http://[::1]/",
+    ):
+        with pytest.raises(ValueError, match="private or loopback"):
+            assert_fetchable_url(hostile)
+
+
 def test_write_and_hash_stay_inside_sandbox(monkeypatch, tmp_path):
     monkeypatch.setenv("S16_SANDBOX_ROOT", str(tmp_path))
     written = write_text_file("reports/result.txt", "verified output")