OWASP Top 10 for LLM Applications 2025
OWASP GenAI Security Project · 2025 · AI
2025 Top 10 Risk & Mitigations for LLMs and Gen AI Apps. Prompt injection remains first (LLM01), then sensitive information disclosure, supply chain, data and model poisoning, improper output handling, excessive agency, system prompt leakage, vector and embedding weaknesses, misinformation, and unbounded consumption. The Station list matches this edition.
British engineering giant Arup revealed as $25 million deepfake scam victim
CNN Business · 2024-05-16 · AI
A Hong Kong finance employee at a multinational engineering firm paid about US$25 million (HK$200 million) across 15 transfers after a video call in which every other participant was synthetic. The company confirmed fake voices and images were used. Hong Kong police described the pattern in February 2024; the company was named in May.
https://www.cnn.com/2024/05/16/tech/arup-deepfake-scam-loss-hong-kong-intl-hnk (opens in a new tab)
FBI and Partners Issue National Public Safety Alert on Financial Sextortion Schemes
Federal Bureau of Investigation · 2022-12-19 · Safety
19 December 2022. Joint alert with Homeland Security Investigations and NCMEC. Over 7,000 reports in the prior year, at least 3,000 minor victims, primarily boys, and more than a dozen suicides.
Sextortion: A Growing Threat Targeting Minors
Federal Bureau of Investigation · 2024-01-16 · Safety
16 January 2024. From October 2021 to March 2023: over 13,000 reports, at least 12,600 victims — primarily boys — and at least 20 suicides. Typical victims males aged 14 to 17. Offenders usually overseas. Digital Safety uses these numbers: at least twenty, not more than twenty.
New FTC Data Show a Big Jump in Reported Losses to Fraud to $12.5 Billion in 2024
Federal Trade Commission (10 March 2025) · 2025-03-10 · Fraud
Consumers reported losing more than $12.5 billion to fraud in 2024, a 25% increase. Investment scams $5.7 billion; imposter scams $2.95 billion. Share of fraud reports with a money loss rose from 27% (2023) to 38% (2024). Sentinel reports are unverified consumer filings.
Consumer Sentinel Network Data Book 2024
Federal Trade Commission · 2025-02-28 · Fraud
6.5 million reports in 2024. Median loss across fraud reports: $497. Data Book aggregates what consumers told Sentinel; it is not a survey and not a verified loss census.
https://www.ftc.gov/reports/consumer-sentinel-network-data-book-2024 (opens in a new tab)
Cryptocurrency and AI Scams Bilk Americans of Billions
Federal Bureau of Investigation · IC3 (6 April 2026) · 2026-04-06 · Fraud
2025 Internet Crime Report: 1,008,597 complaints; nearly $21 billion reported lost. People over 60 reported about $7.7 billion. First IC3 section on artificial intelligence: 22,364 complaints, nearly $893 million. Complaints, not a complete census.
Scammers use AI to enhance their family emergency schemes
Federal Trade Commission Consumer Alert (March 2023) · 2023-03-20 · Fraud
A scammer can clone a relative's voice from a short clip posted online. Do not trust the voice. Hang up and call a number you already know.
FCC Confirms that TCPA Applies to AI Technologies that Generate Human Voices
Federal Communications Commission (FCC-24-17, 8 February 2024) · 2024-02-08 · Fraud
Declaratory ruling: TCPA restrictions on artificial or prerecorded voice cover current AI that generates human voices. Consent required. Makes the act of using AI-generated voices in covered robocalls illegal. Does not make a targeted live clone detectable by ear.
Using Credit Cards and Disputing Charges
Federal Trade Commission Consumer Advice · n.d. · Finance
Federal law limits your responsibility for unauthorized charges to $50; the page does not name the statute (it is 15 U.S.C. § 1643, the next entry). The Fair Credit Billing Act sets the dispute process for billing errors on credit cards and revolving charge accounts: the letter must reach the issuer within 60 days after the first bill with the error was sent. The process does not cover personal loans or loans to buy cars or major appliances.
https://consumer.ftc.gov/articles/using-credit-cards-and-disputing-charges (opens in a new tab)
15 U.S.C. § 1643, Liability of holder of credit card
United States Code (Legal Information Institute, Cornell Law School) · n.d. · Finance
Truth in Lending Act § 133, added by Pub. L. 91-508 on October 26, 1970: a cardholder is liable for unauthorized use of a credit card only if, among other conditions, "the liability is not in excess of $50" and the use occurs before the card issuer has been notified.
https://www.law.cornell.edu/uscode/text/15/1643 (opens in a new tab)
Regulation E § 1005.6 — Liability of consumer for unauthorized transfers
Consumer Financial Protection Bureau · n.d. · Finance
Electronic Fund Transfer Act / Regulation E. Debit and ACH unauthorized transfers: up to $50 if notice within two business days of learning of loss or theft; up to $500 if later; unlimited for later transfers if a statement error is not reported within 60 days. Money leaves the account first. That is the difference from credit.
https://www.consumerfinance.gov/rules-policy/regulations/1005/6/ (opens in a new tab)
Am I responsible for unauthorized charges if my credit cards are lost or stolen?
Consumer Financial Protection Bureau · n.d. · Finance
Report before use: no liability. Unauthorized use before report: $50 federal cap; many issuers waive it. Stolen account number without the card: generally no liability.
Shields Up
Cybersecurity and Infrastructure Security Agency · n.d. · Digital
Federal cyber-defense posture guidance for families, organizations, and corporate leaders: multi-factor authentication, updates, phishing, incident reporting to CISA. Household digital safety, not a guarantee.
https://web.archive.org/web/20260914235715/https://www.cisa.gov/shields-up (opens in a new tab)
Surveillance Self-Defense
Electronic Frontier Foundation · n.d. · Digital
Maintained how-to for protecting communications, built around threat modeling before tools. ShadowPoint teaches the judgment and sends readers here for steps that go stale.
Signal Protocol specifications
Signal Technology Foundation · n.d. · Digital
Open specifications (X3DH, Double Ratchet, PQXDH) and libraries. The protocol is used by Signal and by other messengers. Nonprofit, not an ad-supported product. Taught as what the protocol is, not as a brand endorsement.
Documentation Logs & Victim Handbooks
Stalking Prevention, Awareness, and Resource Center (SPARC) · n.d. · Safety
Incident log as practitioner standard: date, time, place, what happened, witnesses. A pattern is what makes a case. SPARC is a federally funded project of AEquitas; it trains helpers and does not run a victim hotline.
https://www.stalkingawareness.org/documentation-log/ (opens in a new tab)
Take It Down
National Center for Missing & Exploited Children · n.d. · Safety
Free hash-based service to help stop online sharing of nude or sexually explicit images taken of a person under 18. The image stays on the device; only a hash is listed. Adult image help is a different service (StopNCII). CyberTipline remains the reporting route for threats and blackmail.
National Threat Assessment Center
United States Secret Service · n.d. · Safety
Established 1998. Research and guidance for preventing targeted violence. October 2024 guide: Behavioral Threat Assessment Units for state and local law enforcement. Method for a known person of concern over time. Not a tool for assessing strangers.
https://www.secretservice.gov/protection/ntac (opens in a new tab)
Portable generators — carbon monoxide
US Consumer Product Safety Commission (27 May 2026) · 2026-05-27 · Household
Operate generators outdoors only, at least 20 feet from homes, exhaust away from openings. Never indoors or in a garage even with doors open. Look for CO shutoff (PGMA G300-2023 / UL 2201-2023). Working CO alarms on every level.
2025–2026 Winter Reliability Assessment
North American Electric Reliability Corporation (18 November 2025) · 2025-11-18 · Household
Much of North America at elevated risk of insufficient energy in extreme winter. Resources adequate for normal peak; prolonged wide-area cold snaps are the problem. Peak demand up 20 GW since the prior winter. A planning assessment by the reliability organization, not a countdown clock.
https://www.nerc.com/newsroom/2025-winter-reliability-assessment (opens in a new tab)
Community Emergency Response Team (CERT)
Federal Emergency Management Agency · n.d. · Household
Volunteer training in fire safety, light search and rescue, team organization, disaster medical operations. Independent study IS-317.a is the free introduction. Local programs run through fire departments.
CyberTipline
National Center for Missing & Exploited Children · n.d. · Safety
The nation's centralized reporting system for online exploitation of children. Take It Down hashes; CyberTipline is the report.
https://www.missingkids.org/gethelpnow/cybertipline (opens in a new tab)
Carbon Monoxide Information Center
US Consumer Product Safety Commission · n.d. · Household
Federal hub for generator CO guidance and fact sheets. The Commission’s own guidance is to run a generator outdoors, at least 20 feet from the house, with exhaust directed away.
Prompting best practices
Anthropic · n.d. · AI
Brief it like a new colleague, give the reason behind an instruction, use examples, separate material with tags, let it reason first, say what to do instead of what not to do, chain the work, and ground long-document answers in quotes. Vendor documentation.
New AI classifier for indicating AI-written text
OpenAI · 2023-01-31 · AI
The classifier identified 26 percent of AI-written text and mislabeled 9 percent of human text. Withdrawn on July 20, 2023 for its low accuracy.
https://openai.com/index/new-ai-classifier-for-indicating-ai-written-text/ (opens in a new tab)
GPT detectors are biased against non-native English writers
Weixin Liang, Mert Yuksekgonul, Yining Mao, Eric Wu and James Zou · arXiv · 2023-04-06 · AI
Detectors consistently mislabel non-native English writing as AI-generated while native samples are identified correctly. The 61 percent TOEFL figure is in the paper, not the abstract.
NIST SP 800-63B-4, Authentication and Authenticator Management
National Institute of Standards and Technology · 2025-07-31 · Digital
Verifiers shall not impose composition rules, shall not require periodic changes, shall screen new passwords against a list of compromised ones, shall allow password managers and automatic filling, and shall not prompt for security questions.
https://pages.nist.gov/800-63-4/sp800-63b.html (opens in a new tab)
FTC Order Prohibits Data Broker X-Mode Social and Outlogic from Selling Sensitive Location Data
Federal Trade Commission · 2024-01-09 · Digital
Stop selling sensitive location data and delete what was collected. Sensitive locations include medical clinics, places of worship and domestic abuse shelters.
FTC Bans SpyFone and CEO from Surveillance Business
Federal Trade Commission · 2021-09-01 · Digital
Support King (SpyFone) and its CEO are banned from offering any surveillance app or business, must delete the data collected, and must notify owners of monitored devices.
Government Requests
Signal Technology Foundation · n.d. · Digital
The only information Signal can produce under legal process is when an account registered and the last date it connected. It does not hold messages, contacts, groups or profile data.
What To Know About Credit Freezes and Fraud Alerts
Federal Trade Commission Consumer Advice · n.d. · Finance
A freeze costs nothing to place or lift, does not affect your score, and must be placed at all three bureaus. An initial fraud alert lasts a year and needs only one bureau.
https://consumer.ftc.gov/articles/credit-freezes-and-fraud-alerts (opens in a new tab)
Social Media and Adolescent Health, report highlights
National Academies of Sciences, Engineering, and Medicine · 2023-12 · Safety
The literature “did not support the conclusion that social media causes changes in adolescent health at the population level”: small effects, weak associations. Online harassment and bullying are named as harms.
18 US Code § 2511
Legal Information Institute, Cornell Law School · n.d. · Safety
(2)(d): a person not acting under color of law may intercept a communication they are party to, or where one party consented, unless for a criminal or tortious purpose. (4)(a): up to five years. (2)(g)(ii): intercepting radio traffic from a public safety system, including police and fire, that is readily accessible to the general public is not an offense.
https://www.law.cornell.edu/uscode/text/18/2511 (opens in a new tab)
Crime in the United States 2019: Burglary
Federal Bureau of Investigation, Uniform Crime Reporting · 2020 · Household
In 2019, 55.7 percent of burglaries involved forcible entry, 37.8 percent were unlawful entries, and 6.5 percent were attempted forcible entry.
Hurricanes in 2024 led to the most hours without power in the United States in 10 years
US Energy Information Administration · 2025-12-01 · Household
About 11 hours without power per customer in 2024, about 80 percent from major events, mostly hurricanes. Routine interruptions average about two hours a year.
https://www.eia.gov/todayinenergy/detail.php?id=66744 (opens in a new tab)
The February 2021 Cold Weather Outages in Texas and the South Central United States
FERC, NERC and Regional Entity Staff · 2021-12-08 · Household
Freezing and fuel supply problems caused 75.6 percent of unplanned generation outages; 81 percent of freeze-related failures happened above the units' stated design minimums; protecting four component types could have cut ERCOT's outages by an estimated 67 percent (SPP 47, MISO South 55).
IARC Classifies Radiofrequency Electromagnetic Fields as Possibly Carcinogenic to Humans
International Agency for Research on Cancer, Press Release 208 · 2011-05-31 · Household
Group 2B, possibly carcinogenic, on limited evidence for glioma and acoustic neuroma in wireless phone users. Evidence for other cancers judged inadequate.
https://www.iarc.who.int/wp-content/uploads/2018/07/pr208_E.pdf (opens in a new tab)
Cell Phone Radio Frequency Radiation
National Toxicology Program · n.d. · Household
Clear evidence of malignant heart schwannomas in male rats and some evidence of brain gliomas, from whole-body exposure over about two years. NTP cautions against applying the results directly to people.
https://ntp.niehs.nih.gov/whatwestudy/topics/cellphones (opens in a new tab)
Environmental Health Trust v. FCC, No. 20-1025
US Court of Appeals for the D.C. Circuit · 2021-08-13 · Household
The FCC failed to give a reasoned explanation that its guidelines protect against non-cancer effects. “We take no position in the scientific debate regarding the health and environmental effects of RF radiation.”
Carpenter v. United States, 585 US 296
Supreme Court of the United States · 2018-06-22 · Digital
Acquiring historical cell-site location information is a Fourth Amendment search; the government generally needs a warrant.
https://www.supremecourt.gov/opinions/17pdf/16-402_h315.pdf (opens in a new tab)
People v. Goetz
New York Court of Appeals · 1986-07-08 · Safety
Penal Law § 35.15 requires that the defendant actually believed deadly force was necessary and that a reasonable person in his situation could have had those beliefs. The drafters inserted “reasonably” before “believes,” departing from the Model Penal Code's subjective test.
https://www.nycourts.gov/reporter/archives/p_goetz.htm (opens in a new tab)
Effects of Laws Expanding Civilian Rights to Use Deadly Force in Self-Defense on Violence and Crime: A Systematic Review
Yakubovich et al. · American Journal of Public Health · 2021 · Safety
25 studies. Expanding these laws was associated with no change to small increases in violent crime, Florida studies showed 24% to 45% increases in homicide, and outcomes differed by the victim's race. The evidence contradicts claims of deterrence.
https://pmc.ncbi.nlm.nih.gov/articles/PMC7958062/ (opens in a new tab)
Does Strengthening Self-Defense Law Deter Crime or Escalate Violence? Evidence from Castle Doctrine
Cheng Cheng and Mark Hoekstra · NBER Working Paper 18134 (Journal of Human Resources, 2013) · 2012 · Safety
No deterrence: burglary, robbery and aggravated assault were unaffected. Homicides rose by about 8 percent.
Why language models hallucinate
Adam Kalai et al. · OpenAI · 2025-09-05 · AI
Pretraining is next-word prediction. Arbitrary low-frequency facts cannot be predicted from patterns and lead to hallucinations. Standard training and evaluation reward guessing over acknowledging uncertainty, although calibration is achievable.
https://openai.com/index/why-language-models-hallucinate/ (opens in a new tab)
How your data is used to improve model performance
OpenAI Help Center · n.d. · AI
Content in services for individuals may be used for training unless you opt out in Data controls. By default, inputs and outputs from ChatGPT Business, Enterprise, Edu and the API are not used.
Samsung bans use of generative AI tools like ChatGPT after April internal data leak
Kate Park · TechCrunch · 2023-05-02 · AI
After internal data leaked to ChatGPT in April 2023, Samsung restricted generative AI tools on company-owned devices and on non-company devices on its internal networks.
Samsung bans staff from using ChatGPT after data leak
Claudia Glover · Tech Monitor · 2023-05-02 · AI
Staff had entered chip and device source code and internal meeting notes into ChatGPT. The ban on company devices was described as temporary.
https://www.techmonitor.ai/technology/cybersecurity/samsung-bans-chatgpt (opens in a new tab)
LLM06:2025 Excessive Agency
OWASP GenAI Security Project · 2025 · AI
Excessive functionality, permissions and autonomy let prompt injection trigger damaging actions. Limit extensions and permissions to the minimum and require human approval for high-impact actions.
https://genai.owasp.org/llmrisk/llm062025-excessive-agency/ (opens in a new tab)
7z Format
Igor Pavlov · 7-Zip · n.d. · Digital
7z supports AES-256 encryption with a SHA-256-based key derivation. This page does not state that file names are encrypted.
A Known Plaintext Attack on the PKZIP Stream Cipher
Eli Biham and Paul C. Kocher · Fast Software Encryption (FSE 1994), LNCS · 1995 · Digital
The legacy PKZIP cipher's internal key can be found within hours on a personal computer from a few hundred bytes of known plaintext.
https://link.springer.com/chapter/10.1007/3-540-60590-8_12 (opens in a new tab)
#StopRansomware Guide
CISA, MS-ISAC, NSA and FBI · 2023-09 · Digital
Keep offline, encrypted backups of critical data and regularly test them, because many ransomware variants find and delete or encrypt accessible backups.
https://www.cisa.gov/stopransomware/ransomware-guide (opens in a new tab)
Family Radio Service (FRS)
Federal Communications Commission · 2022-09-27 · Household
Licensed by rule, so no individual license. 22 channels in the 462 and 467 MHz range, up to 2 W (0.5 W on channels 8 to 14).
47 CFR § 95.587, Additional requirements for FRS
eCFR · n.d. · Household
The FRS antenna must be a non-removable integral part of the transmitter.
General Mobile Radio Service (GMRS)
Federal Communications Commission · n.d. · Household
A license is required. Ten-year term. Family members may operate within the licensed system, and licensees may use portable, mobile, fixed and repeater stations.
47 CFR § 95.1705, Individual licenses
eCFR · n.d. · Household
An individual GMRS license lasts ten years and covers the licensee's immediate family. No examination is listed.
47 CFR § 95.1763, GMRS channels
eCFR · n.d. · Household
GMRS has 30 channels. The 467 MHz main channels are for control and fixed stations, the repeater inputs.
Multi-Use Radio Service (MURS)
Federal Communications Commission · n.d. · Household
Licensed by rule, five VHF channels near 151 and 154 MHz, no more than 2 W, and an external antenna can extend range.
47 CFR § 95.2741, MURS antenna height limitation
eCFR · n.d. · Household
A MURS antenna may be no more than 18.3 m (60 ft) above ground, or 6.10 m (20 ft) above the structure it is mounted on.
Citizens Band Radio Service (CB)
Federal Communications Commission · 2017-04-13 · Household
No individual license is required. Channel 9 is reserved for emergency communications or traveler assistance.
47 CFR § 97.503, Element standards
eCFR · n.d. · Household
Element 2 (Technician) is 35 questions with 26 needed to pass. The current elements include no telegraphy exam. The page does not give the 2007 date.
NOAA Weather Radio All Hazards
National Weather Service · n.d. · Household
Seven VHF frequencies: 162.400, 162.425, 162.450, 162.475, 162.500, 162.525 and 162.550 MHz.
NWR Specific Area Message Encoding (SAME)
National Weather Service · n.d. · Household
A SAME receiver is set with state and county codes so that it alerts only for the chosen areas.
SKYWARN
National Weather Service · n.d. · Household
SKYWARN is a volunteer program of 350,000 to 400,000 trained severe-weather spotters. Training is free and typically lasts about two hours, and the NWS encourages anyone with an interest in public service to join. The page names no license requirement.
Amateur Radio Emergency Service (ARES)
ARRL, the national association for amateur radio · n.d. · Household
ARES consists of licensed amateurs who have registered their qualifications and equipment with their local ARES leadership for communications duty in the public service when disaster strikes. Every licensed amateur is eligible, whether or not an ARRL member. The page does not describe RACES.
47 CFR § 97.407, Radio amateur civil emergency service
eCFR · n.d. · Household
A station may transmit in RACES only if it is certified by the civil defense organization for the area, and every RACES communication must be authorized by that organization. Permitted traffic is limited to emergency and public-safety messages, public information, and drills or tests of one hour a week, or up to 72 hours twice a year with state approval.
IPAWS Alerting Authorities
Federal Emergency Management Agency · n.d. · Household
Federal agencies and state, local, tribal and territorial governments are eligible to send alerts through IPAWS. Other public or private organizations may also be eligible, depending on their public safety mission.
Sign Up for IPAWS
Federal Emergency Management Agency · n.d. · Household
To become an alerting authority, an organization completes FEMA's IS-247 course, procures IPAWS-compatible software, signs a memorandum of agreement with FEMA, and applies for public alerting permissions, which a designated state official or tribal leadership must review and sign.
Amateur Service Rules (FCC 06-178)
Federal Communications Commission · Federal Register, 72 FR 3081 · 2007-01-24 · Household
Removes the five words-per-minute telegraphy examination from the requirements for the General and Amateur Extra licenses, leaving Elements 2 and 3 for General and 2, 3 and 4 for Extra, and removes paragraph (a) of § 97.503, the telegraphy element. Adopted December 15, 2006; effective February 23, 2007.
Trend toward local police radio encryption grows, as does resistance
Mailyn Fidler · Reporters Committee for Freedom of the Press · 2021-06-14 · Household
At least ten cities and counties encrypted police radio after October 2020, and a California directive required departments to encrypt fully or adopt a hybrid system. News organizations objected, and Colorado legislators planned a bill pushing back against encryption. The piece does not discuss trunked systems.
https://www.rcfp.org/police-radio-encryption-trend/ (opens in a new tab)
Malicious browser extensions impacting at least 3.2 million users
GitLab Threat Intelligence · 2025-02-13 · Digital
A cluster of at least 16 malicious Chrome extensions reached at least 3.2 million users. GitLab assesses that the actor acquired access to at least some of them from their original developers, some of whom stated they were transferring control. Google took all of them down after GitLab's January 2025 report. It advises removing extensions you no longer use and watching for extensions that change permissions or ownership.
Staying Safe with Chrome Extensions
Benjamin Ackerman, Anunoy Ghosh and David Warren · Chrome Security Team, Google · 2024-06-20 · Digital
In 2024 less than 1% of all installs from the Chrome Web Store were found to include malware. An extension found to pose a severe risk is removed from the store and disabled in every browser. Users are told to compare an extension's stated goals with the permissions it requests and the privacy practices it publishes.
Back Up Government Data
Cybersecurity and Infrastructure Security Agency (CISA) · n.d. · Digital
Keep 3 copies of important files, on 2 different types of storage media, with 1 copy stored off-site. Backups are the best hope of recovery from a ransomware attack, and restore procedures should be tested, both fully and partially.
Help: What is “header encryption” exactly?
Igor Pavlov (author of 7-Zip) · 7-Zip forum at SourceForge · 2024-10-20 · Digital
The 7z format keeps all metadata, file names included, in one header. The “Encrypt file names” option in 7-Zip's Add to Archive window and the command-line header-encryption switch do the same thing; with the header encrypted, only its size and checksum stay unencrypted. File-name encryption is an option the user turns on.
https://sourceforge.net/p/sevenzip/discussion/45798/thread/df50f35107/ (opens in a new tab)
7z File Format
Library of Congress · Sustainability of Digital Formats · 2024-04-30 · Digital
Lists AES-256 encryption among the 7z format's main features and says the format also supports file-name encryption. The key is generated from the user's passphrase with SHA-256.
https://www.loc.gov/preservation/digital/formats/fdd/fdd000539.shtml (opens in a new tab)
Optimizing LLMs for Speed and Memory
Hugging Face · Transformers documentation · n.d. · AI
Loading the weights of a model with X billion parameters needs roughly 2 × X GB of VRAM at 16-bit precision, so a 70-billion-parameter model needs about 140 GB. A 15-billion-parameter model measured on the page needed about 29 to 32 GB at 16-bit, about 15 GB at 8-bit and a little over 9 GB at 4-bit. The page says that runs on GPUs that are quite accessible for most people. Quantization trades memory against accuracy and sometimes speed.
https://huggingface.co/docs/transformers/llm_tutorial_optimization (opens in a new tab)
Delaware Code, Title 11, Chapter 4 (§ 464, use of force in self-protection)
State of Delaware · n.d. · Safety
§ 464(a): the use of force upon another person is justifiable when the defendant reasonably believes that such force is immediately necessary to protect the defendant against the other person's unlawful force on the present occasion. The page does not mention the Model Penal Code.
https://delcode.delaware.gov/title11/c004/index.html (opens in a new tab)
47 CFR § 1.1102, Schedule of charges for applications and other filings in the wireless telecommunications services
eCFR · n.d. · Household
The FCC's schedule of application charges lists GMRS among the personal licenses. The personal license table charges $35 for a new license, modification or renewal, and its GMRS row points to the Wireless Telecommunications Bureau Fee Filing Guide for any associated regulatory fee.
Project 25 Frequently Asked Questions, version 1.9
Project 25 Technology Interest Group (PTIG) · 2025-06-03 · Household
PTIG members have identified over 3,368 P25 digital radio systems on the air worldwide, 2,387 conventional and 981 trunked. The FAQ does not discuss scanners or clear versus encrypted operation.
Considerations for Encryption in Public Safety Radio Systems
SAFECOM · Cybersecurity and Infrastructure Security Agency · 2016-09 · Household
Many public safety communications channels are streamed across the Internet and openly broadcast to the public, the media, criminals and potential terrorists, which the fact sheet gives as a reason for agencies to consider encryption. It does not discuss trunked systems.
47 CFR § 1.907, Definitions
eCFR · n.d. · Household
The Universal Licensing System (ULS) is the FCC's consolidated database, application filing system and processing system for all Wireless Radio Services, defined as the radio services authorized in parts 13, 20, 22, 24, 26, 27, 30, 74, 80, 87, 88, 90, 95, 96, 97, and 101.
Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks
Patrick Lewis and others, NeurIPS 2020 · 2020-05-22 · AI
Language models store knowledge in their parameters but have limited ability to access and precisely manipulate it; updating that knowledge and providing provenance for decisions remain open problems. RAG adds a retriever over a document index at answer time.
Building effective agents
Erik S. and Barry Zhang, Anthropic · 2024-12-19 · AI
Defines agents as "systems where LLMs dynamically direct their own processes and tool usage", as distinct from workflows on predefined code paths; agents should get ground truth from the environment at each step and may pause for human feedback at checkpoints. A vendor's own engineering guidance.
https://www.anthropic.com/engineering/building-effective-agents (opens in a new tab)
How YouTube recommendations work
YouTube (How YouTube Works) · n.d. · Digital
YouTube's recommendation system selects videos for each viewer from signals such as viewing behavior, likes, dislikes, subscriptions, and satisfaction surveys. A recommendation reflects the ranking system and the viewer's signals.
https://www.youtube.com/howyoutubeworks/recommendations/ (opens in a new tab)
Measuring the Persuasiveness of Language Models
Anthropic · 2024-04-09 · AI
Of the prompting strategies tested, "the Deceptive strategy, which allowed the model to fabricate information, was found to be the most persuasive overall." A text written to persuade is not thereby accurate.
https://www.anthropic.com/research/measuring-model-persuasiveness (opens in a new tab)
Models overview
Anthropic, Claude Platform documentation · n.d. · AI
Compares the vendor's models on latency, pricing and intelligence, from the slowest and most capable to the fastest and cheapest, and gives each a reliable knowledge cutoff date. The vendor's own description of its models, read on the day accessed.
https://platform.claude.com/docs/en/models/overview (opens in a new tab)
Context windows
Anthropic, Claude Platform documentation · n.d. · AI
The context window is all the text a model can reference when generating a response, "different from the large corpus of data the language model was trained on", a "working memory". Across the turns of a conversation, "previous turns are preserved completely" within it.
https://platform.claude.com/docs/en/build-with-claude/context-windows (opens in a new tab)
Memory FAQ
OpenAI Help Center · n.d. · AI
Saved memories are stored separately from chat history and persist across conversations; deleting the original chat does not delete a saved memory. With chat history referencing on, information from past conversations is used in later ones.
https://help.openai.com/en/articles/8590148-memory-in-chatgpt (opens in a new tab)
FAQ
Ollama documentation · n.d. · AI
"Ollama runs locally. We don't see your prompts or data when you run locally." Cloud-hosted models are different: the service processes prompts and responses to provide them.
A joint checking account owner took all the money out and then closed the account without my agreement. Can they do that?
Consumer Financial Protection Bureau · 2024-12-18 · Finance
"In most circumstances, either person on a joint checking account can withdraw money from and close the account." Terms depend on the account agreement and state law. Date is the page's last review.
A financial planner offered to help with my bill paying and banking, but said the only way she or he could help is if I opened a joint account with both of our names on it. Is this true?
Consumer Financial Protection Bureau · 2024-04-15 · Finance
Joint accounts "usually enable all account holders to withdraw money while both are alive", so "you should be very careful about whose name you add to your bank account." Alternatives such as powers of attorney and trusts can meet the goal without joint access. Date is the page's last review.
Lost or Stolen Credit, ATM, and Debit Cards
Federal Trade Commission Consumer Advice · 2022-01 · Finance
ATM and debit card transactions "take money from your account right away." The page advises: "Carry only the cards you'll need."
https://consumer.ftc.gov/articles/lost-or-stolen-credit-atm-and-debit-cards (opens in a new tab)
Skimming
Federal Bureau of Investigation · n.d. · Fraud
Devices on ATMs, point-of-sale terminals or fuel pumps "capture card data and record cardholders' PIN entries." Criminals "use the data to create fake payment cards and then make unauthorized purchases or steal from victims' accounts."
How To Recover Your Hacked Email or Social Media Account
Federal Trade Commission Consumer Advice · 2023-08 · Digital
Someone who takes over your email "could request a password reset link for any of your other accounts, get the password reset link from your inbox, change your password, and lock you out of the account." Check that the recovery email addresses and phone numbers are ones you entered and can reach.
Sign in to multiple accounts at once
Google Account Help · n.d. · Digital
"If you have more than one Google Account, you can sign in to multiple accounts at once." Vendor documentation of the vendor's own product; not a recommendation of the brand.
https://support.google.com/accounts/answer/1721977 (opens in a new tab)
How to use multiple Proton Accounts in different browser tabs
Proton Support · n.d. · Digital
Proton's web apps can add another account ("Add another account") and keep several accounts open in different browser tabs. Vendor documentation of the vendor's own product; not a recommendation of the brand.
https://proton.me/support/multiple-accounts-browser-tabs (opens in a new tab)
IS-317.a: Introduction to Community Emergency Response Teams
FEMA National Disaster and Emergency Management University · n.d. · Household
About 2 hours. Completion is not equivalent to, and cannot be used in place of, classroom CERT Basic Training.
Non-Fire Carbon Monoxide Deaths, 2022 Annual Estimates
US Consumer Product Safety Commission (May 2026) · 2026-05 · Household
An estimated 892 generator-associated non-fire CO deaths from 2012 through 2022, 41 percent of consumer-product CO deaths, including 93 in 2022.