A(i). Policy 2 forbids reusing a library password on any personal account. This protects staff because when an unrelated site is breached and its password list is leaked, that stolen password will not open the catalog system, so a credential-stuffing attempt fails. (Policy 5, capping editor rights at two staff per branch, is also acceptable: it limits how many accounts an adversary can target to reach the patron database.)

A(ii). Policy 1 could be strengthened by requiring multifactor authentication in addition to the password. For example, requiring a one-time code from an authenticator app means the credentials captured in Source 1 would not have been enough to sign in at 02:07 on March 11. (Also acceptable: rewrite Policy 3, since a padlock only proves the connection is encrypted, not that the site is genuine, so it should require a VPN or forbid catalog access from open networks entirely.)

B(i). Spear phishing. (Pretexting is also acceptable if the evidence in B(ii) supports it, since the message invents a migration story to justify the request.)

B(ii). The message is aimed at one named person using details specific to her. It addresses Dana by first name, references Monday’s branch meeting, and states that she is one of only two staff with editor rights, which matches Policy 5. It also creates urgency with a 5:00 p.m. deadline and the threat of a three-week suspension. The sender domain rivertonl1b.org substitutes the digit 1 for the letter i in the real domain, and the displayed link text catalog.rivertonlib.org/login does not match the actual target rivertonl1b-secure.net/verify.

C(i). Two access points broadcast the same SSID, ThirdStreetCafe, with different MAC addresses. The legitimate one, A4:2B:8C:11:07:D1, uses WPA2 and has been seen since March 02. The second, 66:F0:19:4E:B2:07, is open with no password, appeared for the first time on March 10, and has a much stronger signal at -38 dBm. A duplicate SSID that is newer, open, and closer is an evil twin.

C(ii). All traffic from a device on that access point passes through equipment the adversary controls, putting them in the middle of the connection. They can present a fake catalog sign-in page, or strip the connection down to unencrypted HTTP, so the username and password Alvarez types are captured before the traffic ever reaches the real catalog server. Because the network is open, there is no encryption between her device and the access point at all.

C(iii). Use a VPN so all traffic is encrypted end to end even on an untrusted network, which means an evil twin sees only ciphertext. (Also acceptable: use a phone’s cellular hotspot instead of public Wi-Fi; verify the network with staff before connecting; disable automatic reconnection to open networks.)

D(i). Row 3. From row 3 onward the source address is 192.0.2.181 with an unknown geolocation and a client string of python-requests, while every genuine Alvarez session uses a Riverton address and Chrome on Windows. The 02:07 timestamp is also far outside her working hours. Row 3 is the adversary’s first attempt; row 5 is where they succeed.

D(ii). Row 6 modifies 1,204 rows in the patrons table, so the integrity of the patron records is destroyed; the library can no longer trust that any patron record is accurate. Row 7 exports all 1,204 rows, so confidentiality is lost as well: the adversary now holds a copy of the personal data of every patron in that table.

D(iii). A brute-force attack is many rapid guesses working through a large number of candidate passwords. Here there are only two failures before success, one second apart. That pattern fits an adversary who already holds the credentials and is retrying a small error, which matches the phishing capture in Source 1, rather than guessing the password from scratch.

E(i). The system holds a baseline of Alvarez signing in between 07:30 and 18:00 from Riverton using Chrome on Windows. It observed a 02:07 sign-in, an unrecognized network, and a python-requests client, and it flagged a bulk export of 1,204 rows against a baseline maximum of 40. Each observation is a departure from the learned pattern, which is why confidence reached 0.94.

E(ii). The ACTION line states that the alert was written to a queue and that no automatic account action was configured. Detection and response are separate things: the system correctly recognized the anomaly at 02:08, one minute before the record edit at 02:09, but it was only able to record a notice for a human to read later. With no analyst on duty at 2 a.m., nothing stopped the session.

E(iii). Configure the system to act on a high-confidence alert instead of only logging it: automatically suspend the session and require re-authentication with a second factor, or block the source address and freeze bulk exports on the account until a human reviews it. At 0.94 confidence and one minute before the first edit, an automated session termination would have prevented both the edit and the export.

E(iv). Automated action on a model’s judgment produces false positives. A librarian legitimately working late, travelling, or running an approved end-of-year export could trip the same flags and be locked out of the catalog with no way to work and no analyst available to reverse it. Aggressive automation trades availability for confidentiality and integrity, so the confidence threshold and the severity of the action have to be tuned deliberately.