Skip to content
IAM Insights

Explainers · Glossary

LLM

large language model · large language models

The text-predicting AI model behind chatbots and agents. It follows instructions statistically, which is why an instruction hidden in its input can redirect it.

In our briefings

The claim each of the newest briefings on LLM would have you leave with.

  1. Mattson, SecureAuth's chief executive, argues that a model cannot be secured the way tested software is: it is non-deterministic, and a request phrased another way, or in another language, gets around whatever it was told not to do. His advice is to "just assume that it could do anything any time" and put the guardrails around the model, so that a financial transaction or a database change the agent attempts is approved and recorded. He cites Anthropic as saying no guardrail is adequate. The statement we take him to mean, Anthropic's of 12 June, which he recalls as "three weeks ago, I believe", says the company suspects perfect jailbreak resistance "is not currently possible for any model provider", while defending its own safeguards as strong. Briefings we have published on other sources move the control off the model too: a Silverfort guest calls guardrails on a probabilistic system guidance rather than control, ClearVector's John Laliberte compares them to bowling bumpers, and a zero-trust framework paper for agents adds a session-enforcement layer.

    When the model cannot be trusted, control what the agent can reach, checked on every actionBI-FIND-2026-0151

  2. GitHub Security Lab's Man Yue Mo and Peter Stöckli at first had the model gather facts such as a workflow's triggers from the source code, through instructions in its prompts. That worked most of the time, they write, but the model sometimes recorded only some of the events that trigger a workflow, or reached inconsistent conclusions about whether a trigger runs it in a privileged context. They moved most of the work a program can easily do into tools on their MCP servers and left the model the harder reasoning, such as finding permission checks, which they say led to "a much more consistent outcome". They also split triage into small tasks that each start with a fresh context and store their results in a database, so a failed run restarts at the failed task and reuses the earlier results.

    GitHub's AI triage agent got more consistent once code, not the model, ran the easy checksBI-FIND-2026-2407

  3. AP2 v0.2 authorizes an agent's purchase with two signed objects, a Checkout Mandate and a Payment Mandate, both linked by hash to the merchant's own checkout. The Checkout Mandate carries a checkout_hash claim that binds it to a Checkout JWT the merchant signed, and the Payment Mandate is bound to the hash of that same JWT. Both travel as SD-JWTs, a JWT format with selective disclosure, whose type claim carries a version, such as mandate.payment.1 or mandate.checkout.open.1, and implementations must match it exactly. The merchant, the credential provider and network, and the payment processor each have checks the spec says they must run, and any validation or processing the spec assigns to a role must run in deterministic code, whether or not that role is agentic. The merchant's Checkout JWT must not use a deterministic signature such as Ed25519; the spec gives ECDSA as the example and says the rule is there to prevent rainbow-table attacks. Whatever mode produced them, verifiers always receive closed mandates.

    Google's AP2 v0.2 ties an agent's purchase to the merchant's own signed checkout by hashBI-FIND-2026-0414

›Cited in 127 briefings

Every briefing that discusses this term, newest first.

  1. When the model cannot be trusted, control what the agent can reach, checked on every action

    Podcast · · Tech Talks Daily · AI Agent Security: Why Identity and Access Control Matter More Than Guardrails · Aug 11, 2026

  2. GitHub's AI triage agent got more consistent once code, not the model, ran the easy checks

    Article · · The GitHub Blog · Man Yue Mo et al. · Jan 20, 2026

  3. Google's AP2 v0.2 ties an agent's purchase to the merchant's own signed checkout by hash

    Article · · ap2-protocol.org · Google · Apr 28, 2026

  4. A new STIR draft would let a number's holder require AI agents calling from it to say so

    Draft · · IETF Datatracker · Jon Peterson, TransUnion · Nov 2, 2026

  5. Palo Alto's 109-to-1 machine identity ratio is mostly AI agents, and cannot be read as last year's 82 grown

    Article · · Palo Alto Networks · Amy Blackshaw et al. · May 14, 2026

  6. Google's bug-hunting agent found hundreds of XSS bugs and none on its hardened stack, says its web-security lead

    Video · · OWASP Foundation · Uploaded October 24, 2026

  7. A user's approved request is not yet authority an agent can be held to, and most deployments skip that step

    Article · · Control Plane by Karl McGuinness · Karl McGuinness · Mar 17, 2026

  8. Enforced SMB and LDAP signing did not stop NTLM relay until Microsoft's September and October 2026 patches

    Podcast · · Risky Business · #859 · Nov 11, 2026

  9. Woven by Toyota's exception agent recommends; a person decides, because a human has to answer for it

    Video · · Agentic AI Foundation · Uploaded November 4, 2026

  10. RBC's head of AI says most of its 50,000-plus agents are approved straight through

    Podcast · · Tech Disruptors · RBC on Turning AI Into a Competitive Advantage · Oct 20, 2026

  11. ceLLMate checks a browser agent's HTTP requests, not its clicks, and blocked all 12 scripted GitLab attacks

    Paper · · arXiv · Luoxi Meng et al. · Dec 14, 2025

  12. Apps built by the same code model repeat its security flaws, and their visible features hint at which ones

    Video · · Black Hat · Uploaded November 3, 2026

  13. NatWest found its LLM judge barely tracked human reviewers, and an 83% pass rate hid 43% on topic changes

    Video · · AI Engineer · Uploaded November 4, 2026

  14. Malware now tells AI triage tools to ignore it, and in Talos's test the bluntest request worked best

    Article · · Cisco Talos Blog · Ryan Fetterman · Oct 8, 2026

  15. WIMSE's agent draft adds how an agent's access ends and where policy engines fit, without a new requirement

    Draft · · IETF WIMSE Working Group · Pieter Kasselman et al. · Sep 15, 2026

  16. Raut's revised agent draft keeps token exchange out of the model's reach, and concedes the token carries no evidence of it

    Draft · · IETF Datatracker · Ashay Raut · May 22, 2026

  17. Each agent and each tool call can pass its check while the system fails, an Oxford-led survey argues

    Paper · · arXiv · Christian Schroeder de Witt · Apr 29, 2026

  18. LinkedIn, Uber and State Farm already run SPIRE for workload identity, and each plans to extend it

    Field signals · · 12 voices from the field · Nov 2024 to Oct 2026

  19. A point-in-time access review can't keep pace with changing access, four voices say, and four vendors now put AI inside the review

    Field signals · · 9 voices from the field · Nov 2024 to Aug 2026

  20. Shai-Hulud v4 used an MCP tool description to tell AI assistants to read SSH keys and keep quiet about it

    Video · · OWASP Foundation · Uploaded October 22, 2026

  21. Mozilla's AI bug hunting got past hallucinated reports once a harness checked each finding mechanically

    Video · · OWASP Foundation · Uploaded October 21, 2026

  22. IBM Research puts agent identity in a SPIFFE and OAuth sidecar; Maia Iyer calls authorization policy still open

    Podcast · · Kubernetes Bytes · Your AI Agent Has Too Much Access | Fixing Agent Identity · Sep 15, 2026

  23. The OWASP MCP Top 10 lead says few security teams can find the MCP servers developers run on their own machines

    Podcast · · The Application Security Podcast · Shadow MCP Servers: The AI Security Risk Hiding on Your Developers' Laptops · Oct 12, 2026

  24. SaaS apps that only know users end up giving agents human accounts, says Okta's former chief architect

    Podcast · · The Identity Jedi Show: Where Identity Meets Strategy · Agents need a mission, not a key. · Oct 21, 2026

  25. Mythos also reports bugs that are not there, says Trend AI's CISO, so the work is in checking them

    Podcast · · Cloud Security Podcast · Mythos Hype vs. Reality: Why You Need an Agentic Harness · Oct 21, 2026

  26. Agents exceeded the user's authorization in 19.2% of a benchmark's runs, more under benign pressure than attack

    Paper · · arXiv · Zhiping Zhang et al. · Oct 20, 2026

  27. Of 30 deployed AI agents, 7 publish how a website can recognize them and one signs its requests

    Paper · · arXiv · Leon Staufer et al. · Feb 19, 2026

  28. Orange researchers would drop certificates from agent handshakes by putting post-quantum keys on a ledger

    Paper · · arXiv · Karol Budzyn et al. · Oct 16, 2026

  29. Six companies' agent-auth draft, since adopted by the WIMSE group, keeps credentials out of the model's reach

    Draft · · IETF Internet-Draft (individual submission, since adopted by the WIMSE working group) · Pieter Kasselman et al. · Jul 6, 2026

  30. A quarter of 31,132 agent skills matched a risky pattern; 23 of the 87 riskiest showed clear signs of malice

    Paper · · arXiv · Yi Liu et al. · Jan 15, 2026

  31. AgentArmor cut attack success to 3% in its authors' tests; most misses came when a user said to follow a file

    Paper · · arXiv · Peiran Wang et al. · Nov 18, 2025

  32. Sysdig's AI-run ransomware case used old tricks: default logins, stored keys, a root password of unknown origin

    Article · · Sysdig · Michael Clark · Jul 1, 2026

  33. With real MCP tools for cloud, database and email, most injected attacks got past four models, even with a plan

    Paper · · arXiv · Quan Zhang et al. · Mar 30, 2026

  34. A phishing kit aimed at Taiwan researchers relays each Google sign-in step and gives the attacker the session

    Article · · Cisco Talos · Joey Chen · Oct 8, 2026

  35. A Windows credential stealer lets four commercial AI models vote on its next move, choosing from four options

    Article · · Cisco Talos · Ryan Fetterman · Sep 22, 2026

  36. OWASP's agent Top 10 blames identity systems built for people, and does not say how it chose or ranked its risks

    Article · · OWASP GenAI Security Project · John Sotiropoulos et al. · Dec 9, 2025

  37. NIST's agent identity work starts from OAuth, SPIFFE and SCIM and defers agents from outside the enterprise

    Article · · NIST National Cybersecurity Center of Excellence · Harold Booth et al. · Feb 5, 2026

  38. Saviynt's strategy head calls agent discovery a commodity and pitches Zuma for runtime control

    Podcast · · The Non-Human & AI Identity Podcast · Ep # 14 - Turning an AI Breach into a Case Study · Sep 15, 2026

  39. In Accenture's agent-delegation design, no outside party checks the user, and revocation is left open

    Paper · · arXiv · Srikumar K. Subramanian et al. · Sep 25, 2026

  40. PayPal moved MCP tool permission checks into the server, and prompt-injected agents stopped getting through

    Paper · · arXiv · Huan Li et al. · Sep 18, 2026

  41. Microsoft's Access Review Agent suggests whether to keep or remove access; the reviewer still decides, in Teams

    Podcast · · Entra.Chat · AI is Coming to Identity Governance! Meet the Entra Access Review Agent · Oct 18, 2025

  42. WorkOS's Tobin South, in November 2025: agents will get a new kind of account, not a new identity standard

    Podcast · · CodingCat.dev Podcast · 5.10 - How to Secure AI - Identity Access Management for Agents · Nov 21, 2025

  43. Four vendors now say the access review is finished for agents. This one adds that agents will approve agents.

    Podcast · · The Non-Human & AI Identity Podcast · Ep # 15 - Identity is the 3rd Generation of Security · Sep 22, 2026

  44. The 'intent' in the Agentic JWT draft is a workflow step, and the agent reports its own fingerprint

    Draft · · IETF Datatracker · Abhishek Goswami · Jul 1, 2026

  45. The IBAC paper's strict mode blocked every prompt injection, and its agent finished only a third of its tasks

    Article · · ibac.dev · Jordan Potti · Mar 3, 2026

  46. MCP made authorization optional, and two identity veterans say most servers took the option

    Podcast · · Hybrid Identity Protection Podcast · Agentic AI and the Authorization Gap No One Closed · Jun 9, 2026

  47. Two ways to give an AI agent an OAuth identity clashed in April; a draft now lets them work together

    Podcast · · Hybrid Identity Protection Podcast · Securing Non-Human Identities in the Age of Agentic AI with Sarah Cecchetti, Director of Product Management at Semperis · Apr 28, 2026

  48. SecureAuth's sales chief sells agent identity on an unsourced ratio and on sums buyers work out from their own headcount

    Podcast · · GTM Vault · Securing AI Agents Beyond Authentication | Mark van Oppen, SecureAuth · Jul 5, 2026

  49. Mission-Bound OAuth's author now says the mission object does not belong in OAuth

    Article · · Control Plane by Karl McGuinness · Karl McGuinness · Mar 15, 2026

  50. Ping moved agent identity from roadmap to shipping product, and let AI agents run its admin tools too

    Conference · · Ping Identity · Austin · 2-day event (Day 1 / Day 2), 2026

  51. Agent identifiers are split across JWT-based, SPIFFE, WIMSE and proprietary schemes, an identity lead says

    Podcast · · Identity Decoded · Winning in securing agentic identity: Start with one team, one use case · Aug 12, 2026

  52. Once people hand their credentials to an AI, the question is whether the account is still human

    Podcast · · Nexus: A Claroty Podcast · John Laliberte on AI and Non-Human Identity Controls · Aug 5, 2026

  53. Canva sandboxes bounded conversion jobs; broad workforce agents pose a different access problem

    Podcast · · AI Security Podcast · Why I Dont Trust Your AI Agent | Kane Narraway, Canva · Aug 20, 2026

  54. Agent payment protocols prove an agent may pay; a Toronto paper adds a step to check the job was done

    Paper · · arXiv · Adrian de Valois-Franklin et al. · Jun 7, 2026

  55. Putting AIRGuard's policy in an agent's prompt barely cut attacks; enforcing it before each tool call did

    Paper · · arXiv · Suliu Qin et al. · May 27, 2026

  56. The same model took out-of-scope actions in 1.1% of runs in OpenHands and 27.7% in Claude Code

    Paper · · arXiv · Yubin Qu et al. · May 18, 2026

  57. Clicks and timing alone told Oxford researchers which of 14 models was driving a browser agent

    Paper · · arXiv · William Lugoloobi et al. · May 14, 2026

  58. One x402 payment, replayed 1,000 times, was accepted up to 248 times; no audited SDK ties it to what it paid for

    Paper · · arXiv · Zelin Li et al. · May 12, 2026

  59. A forged invoice gets past an auditor that traces which text drove each AI agent action

    Paper · · arXiv · Shihao Weng et al. · May 5, 2026

  60. AgentDID checks an AI agent's identity and readiness in 13.5 seconds, but was tested on at most 50 agent pairs

    Paper · · arXiv · Minghui Xu et al. · Apr 28, 2026

  61. AITH claims 4.7 million agent checks a second, but the test simulated its post-quantum signatures

    Paper · · arXiv · Zhaoliang Chen · Apr 9, 2026

  62. In a stress test, Claude Code made changes through file edits that auto mode's permission check never sees

    Paper · · arXiv · Zimo Ji et al. · Apr 4, 2026

  63. A warning in the system prompt did not reliably stop coding agents obeying lines planted in a skill file

    Paper · · arXiv · David Schmotz et al. · Feb 23, 2026

  64. An LLM pipeline found 45% of LinkedIn users' Hacker News accounts at 99% precision; an older method found 0.1%

    Paper · · arXiv · Simon Lermen et al. · Feb 18, 2026

  65. MACAW Security's trust layer signs every agent call, but what stopped its prompt-injection test was a policy

    Paper · · arXiv · Mohan Rajagopalan et al. · Feb 11, 2026

  66. Auth0's AgentCore guide says prompts alone are not enough, then lets the prompt decide when to ask a person

    Article · · Auth0 Blog · Kapil Patil · Feb 10, 2026

  67. A third-party agent talked a trusted one into unlocking the door; SEAgent blocks it, given the right labels

    Paper · · arXiv · Zimo Ji et al. · Jan 17, 2026

  68. On test pages, lures steered open-source web agents in 67.5% of tasks; a task-scoped check cut that by 78.1%

    Paper · · arXiv · Xinyi Wu et al. · Jan 12, 2026

  69. Sola Security graded its own identity-posture agent: 0.95 on AWS, 0.65 on Okta, and no rival to compare

    Paper · · arXiv · Gal Engelberg et al. · Jan 11, 2026

  70. Off-the-shelf AI agents tied six of 24 scientist interviews in Anthropic's redacted dataset to specific papers

    Paper · · arXiv · Tianshi Li · Jan 9, 2026

  71. A Shenzhen paper ties each AI agent to a verified owner and its exact code, at 15 to 40 seconds of proving a step

    Paper · · arXiv · Zibin Lin et al. · Dec 19, 2025

  72. Refuse the risky action when its plan relies on untrusted input, a Beihang survey of agent attacks proposes

    Paper · · arXiv · Guanquan Shi et al. · Dec 7, 2025

  73. In LLM agent tests, an attacker next to the agent holding private data got 21.48% of it out; five hops away, 1.28%

    Paper · · arXiv · Jinbo Liu et al. · Dec 4, 2025

  74. No app, no developer: least privilege for an AI agent has to come from the task, six security researchers argue

    Paper · · arXiv · Mihai Christodorescu et al. · Dec 1, 2025

  75. CORAL's public agent server rejected forged messages but let a wrong key open the session stream

    Paper · · arXiv · Yedidel Louck et al. · Nov 5, 2025

  76. On its authors' tasks, an LLM check of agent scopes works at one tool and refuses most valid three-tool requests

    Paper · · arXiv · Majed El Helou et al. · Oct 30, 2025

  77. A "don't ask again" for Python let a tampered Agent Skill upload slides from Claude Code

    Paper · · arXiv · David Schmotz et al. · Oct 30, 2025

  78. AAGATE maps NIST's AI risk framework onto Kubernetes, and its paper never times the 'millisecond kill switch'

    Paper · · arXiv · Ken Huang et al. · Oct 29, 2025

  79. A decade of AI privacy papers: nearly half on training-data leakage, 2% on agents leaking user context

    Paper · · arXiv · Niloofar Mireshghallah et al. · Oct 2, 2025

  80. Agentic JWT binds an agent's token to its task and its own code, but it is one author's draft, not a standard

    Paper · · arXiv (cs.CR preprint 2509.13597) · Abhishek Goswami · Sep 16, 2025

  81. A guide to securing agent plans has the planner choose each step's tools and leaves out agent identity

    Paper · · arXiv · Ron F. Del Rosario et al. · Sep 10, 2025

  82. Tool poisoning runs through calls an agent is allowed to make; MindGuard looks for it in the model's attention

    Paper · · arXiv · Zhiqiang Wang et al. · Aug 28, 2025

  83. Twenty models nearly always name real MCP tools; the judge splits them on planning, which a schema check cannot catch

    Paper · · arXiv · Zhenting Wang · Aug 28, 2025

  84. Anthropic says Claude let North Korean operators who cannot code unaided pass interviews at US tech firms

    Article · · Anthropic · Anthropic threat intelligence team · Aug 27, 2025

  85. MCPTox: a poisoned tool description steers most agents into misusing the tools they already hold

    Paper · · arXiv · Haohua Du · Aug 19, 2025

  86. MCP-Guard's classifier alone came within half a point of its best full pipelines' F1 on the authors' test set, and ran faster

    Paper · · arXiv · Wenpeng Xing et al. · Aug 14, 2025

  87. MI9 re-checks agent permissions when goals change; its 99.81% score comes from tests Gemini wrote and scored

    Paper · · arXiv · Charles L. Wang et al. · Aug 5, 2025

  88. A national-lab provenance model logs an AI agent's prompts and decisions, but not whom it acted for

    Paper · · arXiv · Renan Souza et al. · Aug 4, 2025

  89. An MCP privilege study says obscure servers are riskier; its own table points the other way

    Paper · · arXiv · Zhihao Li et al. · Jul 5, 2025

  90. Meta's injection-trained Llama kept agent attacks to 2.1% or less, but its authors tried no attack tuned against it

    Paper · · arXiv · Sizhe Chen et al. · Jul 3, 2025

  91. The July 2025 Cross App Access draft lets the IdP decide app-to-app access and treats agents as plain clients

    Draft · · IETF Datatracker · Aaron Parecki et al. · Jul 2, 2025

  92. RAS-Eval cites weak MCP authentication, yet maps none of its 3,802 agent attacks to an authentication weakness

    Paper · · arXiv · Yuchuan Fu et al. · Jun 18, 2025

  93. Each of eight AI models fell for at least one jailbreak planted in a simulated MCP server

    Paper · · arXiv · Junfeng Fang et al. · Jun 16, 2025

  94. An agent's plan fixed before it reads untrusted data still leaves each call's arguments open to injection

    Paper · · arXiv · Luca Beurer-Kellner et al. · Jun 10, 2025

  95. Customers will bring their own agents, and the enterprise cannot control them: Ping's product chief, June 2025

    Podcast · · Crucial Tech · Episode 10:22 Ping and Identity Access Management in the AI age · Jun 4, 2025

  96. In Agent Security Bench, eight text-level defenses left agent hijacking high; none checks the tool call

    Paper · · arXiv · Hanrong Zhang et al. · May 30, 2025

  97. In Microsoft's agent tests, label checks blocked policy-breaking tool calls and its FIDES planner finished more tasks

    Paper · · arXiv · Manuel Costa et al. · May 29, 2025

  98. AgentDNS gives AI agents one sign-in by having a central root hold each vendor's key

    Paper · · arXiv · Enfang Cui et al. · May 28, 2025

  99. Calling an MCP tool 'the best tool in the world' won the model's pick in almost every test

    Paper · · arXiv · Zihan Wang et al. · May 16, 2025

  100. Decentralized AI agents have no off switch by default; builders say cutting API access still stops most of them

    Paper · · arXiv · Botao Amber Hu et al. · May 14, 2025

  101. Meta's LlamaFirewall cut agent attacks to 1.75% in a replay that counts late catches as defended

    Paper · · arXiv · Sahana Chennabasappa et al. · May 6, 2025

  102. SAGA meters agent-to-agent contact through one registry; its speed and 300-million-agent figures are modelled

    Paper · · arXiv · Georgios Syros et al. · Apr 27, 2025

  103. MCP Guardian wraps tool calls in token and regex checks, leaving poisoned servers and agent IDs open

    Paper · · arXiv · Sonu Kumar et al. · Apr 17, 2025

  104. In a 2025 MCP audit, Claude refused an encoded backdoor but wrote the plain one into .bashrc

    Paper · · arXiv · Brandon Radosevich et al. · Apr 2, 2025

  105. Google's CaMeL stops nearly all prompt injection in tests, but agents finish fewer tasks and need written policy

    Paper · · arXiv · Debenedetti · Mar 24, 2025

  106. Splitting an agent's privileges stopped every tested prompt injection, if users refuse each alert

    Paper · · arXiv · Juhee Kim et al. · Mar 17, 2025

  107. All eight prompt-injection defenses in this test were bypassed by attacks tuned to each one

    Paper · · arXiv · Qiusi Zhan et al. · Feb 27, 2025

  108. An attacker on one link between AI agents steered the system in over 40% of runs in each of the paper's main tests

    Paper · · arXiv · Pengfei He et al. · Feb 20, 2025

  109. Testing AI agents one at a time misses attacks that combine them, a Cooperative AI Foundation report argues

    Paper · · arXiv · Lewis Hammond · Feb 19, 2025

  110. Filtering what an AI agent sees cut leaks to 2% or less in a simulation, but in-budget upselling got through

    Paper · · arXiv · Sahar Abdelnabi et al. · Feb 3, 2025

  111. An MIT-led paper has the user sign the agent's delegation token and lets OpenID Connect vouch for both parties

    Paper · · arXiv · Tobin South et al. · Jan 16, 2025

  112. CSIRO's agent-tracing taxonomy has fields for tools and guardrails, none for who the agent acted for

    Paper · · arXiv · Liming Dong et al. · Nov 8, 2024

  113. A planted prompt can copy itself from agent to agent until it reaches one that can send data out

    Paper · · arXiv · Donghyun Lee et al. · Oct 9, 2024

  114. A Wisconsin agent defense stopped every benchmark injection by keeping all tool output from its planner

    Paper · · arXiv · Fangzhou Wu et al. · Sep 27, 2024

  115. A 2024 USENIX benchmark found none of ten prompt-injection defenses sufficient, its best detector included

    Article · · USENIX Security 2024 · Yupei Liu et al. · Aug 14, 2024

  116. AgentDojo's best defence against prompt injection was a tool list fixed before the agent reads anything

    Paper · · arXiv (NeurIPS 2024 Datasets and Benchmarks) · Edoardo Debenedetti et al. · Jun 19, 2024

  117. One hijacking question cut a Gemini Ultra agent's privacy score to 33.1%; choosing its data first kept 93.2%

    Paper · · arXiv · Eugene Bagdasarian et al. · May 8, 2024

  118. OpenAI's instruction hierarchy ranks who may give orders by where the text arrived, not by who sent it

    Paper · · arXiv · Eric Wallace et al. · Apr 19, 2024

  119. IsolateGPT walls off each app an AI assistant uses, and its attack results assume the user refuses every warning

    Paper · · arXiv · Yuhao Wu et al. · Mar 8, 2024

  120. GPT-4 obeyed planted instructions more often than any of 24 lower-rated models in the BIPIA test

    Paper · · arXiv (KDD 2025) · Jingwei Yi et al. · Dec 21, 2023

  121. The 2023 paper that named indirect prompt injection showed an attacker can use whatever access the assistant has

    Paper · · arXiv (cs.CR) · Kai Greshake et al. · Feb 23, 2023

  122. The 2022 'ignore previous prompt' paper hijacked GPT-3 58.6% of the time, with no tools or agent in the test

    Paper · · arXiv · Fábio Perez et al. · Nov 17, 2022

  123. ReAct's agents searched Wikipedia and a test store; the authors limited risk by offering no dangerous actions

    Paper · · arXiv · Shunyu Yao et al. · Oct 6, 2022

  124. In 2025, Kuppinger said Microsoft was underestimated and expected EU wallets to make verification reusable

    Podcast · · Identity at the Center · #326 - IAM in 2025 with Martin Kuppinger · Jan 20, 2025

  125. In 2024, KuppingerCole asked for per-task, short-lived agent access without saying how to issue it

    Article · · KuppingerCole Analysts · Anne Bailey · Dec 17, 2024

  126. Microsoft's 2023 keynote: workload standards should carry identity across service calls first, policy later

    Video · · Identiverse - A CRA Resource · Recorded May 31, 2023 · Uploaded June 17, 2024

  127. Ping's product architect in May 2024: ForgeRock's cloud and PingOne both stay after the merger

    Podcast · · Identity at the Center · #279 - AI in IAM with Patrick Harding of Ping Identity · May 6, 2024

Related terms

What appears alongside LLM in the same briefings.

IAMMachine learningLeast privilegeNon-human identitySSOAndromeda SecurityMFAJIT