Debra Baker

The rule repeated across posts: give an agent its own login, limit what it can do, and make sure it can be shut off in one step.

• The follow-on becomes an audit question software sellers should expect from enterprise buyers and SOC 2 auditors — can you tell agent traffic from human traffic?
• For defense contractors the same problem lands under the 72-hour cyber-incident reporting rule in DFARS 252.204-7012, which, as Baker puts it, does not wait while a company works out what its AI agent did.

Relaying reporting that major U.S. banks are detecting third-party AI agents inside customer accounts, Baker notes that Morgan Stanley has begun blocking third-party agents while JPMorgan, Bank of America and Citi are restricting them, and includes Morgan Stanley's statement that it has not observed rogue agents on its platforms. In a separate post Baker walks through an unconfirmed report of an Accenture contractor connecting a personal Claude Code agent to a secure defense network — noting Accenture calls the claims without merit and that none of it has been independently confirmed — then lists three questions: which AI tools are approved for systems touching controlled unclassified information, do agents get scoped revocable access, and does the incident response plan cover an AI tool sending data somewhere it shouldn't.

An earlier post makes the general case that founders underestimate how AI audits affect operational security and data integrity; the later posts turn that into specific credential-scoping and 72-hour-reporting questions.

Read original · Original 2 · Original 3

On SOC 2, she draws the AICPA line between Type 1 (are controls designed right, as of one date) and Type 2 (did they operate over a period, including the auditor's tests and results), argues that enterprise buyers want proof controls work over time, and concludes it is not about having a report but about having the report the buyer will accept.

• The recommended move is to email your biggest prospect and ask which report they need — one email can save an audit.
• The same instinct shapes the CMMC advice: a C3PAO certificate lets a contractor show a prime proof rather than a promise, so ask the prime directly whether it will want a certificate.

The CMMC post notes that Level 2 C3PAO assessments are suspended so contractors self-attest, that C3PAOs are still running Level 2 certification assessments, and that the Cyber AB counted 2,362 final Level 2 certificates as of September.

Read original · Original 2

Contractors currently self-attest, and the things that still apply are the SPRS score, the annual affirmation, DFARS 7012 and NIST SP 800-171.

• The sharpest point is the consequence: an inflated SPRS score is a False Claims Act risk.
• The takeaway she gives is that when you attest, make sure it is true.

The post reads as a short warning built from four bullet points of continuing obligations, closing with the line that self-attesting is a statement to the government and the instruction to get a second set of eyes before signing.

Read original

She frames these as the contents of a board slide and converts them into three questions for an IT provider, including whether at least one backup is offline and when a restore was last tested.

Baker quotes a guest separately: Jason Clinton, Anthropic's deputy CISO, posted on October 22 asking security leaders to go to their boards and ask for help right now — "not for ai, for the basics."

• She then uses the same week's joint FBI, CISA and international advisory on SafePay ransomware (AA26-293A) to show why: at least 600 victims since August 2024, entry through exposed or guessed passwords, VPN access and five known Cisco, Veeam and SonicWall flaws from 2020 to 2024, and attacks on recovery that stopped shadow copies and wiped backup storage.
• Baker's own closing line, after more than 30 years in security, is that the basics are not the boring part of the plan — they are the plan.

Read original

Bringing a security advisor into the renewal process means verifying MFA, the incident response plan and the backup architecture before you need to rely on the policy. The second post names where coverage can be thinner than buyers assume: sub-limits, waiting periods and exclusions, with social engineering fraud, business interruption and nation-state-related incidents called out as areas with meaningful limitations, and a recommendation to review the fine print with a broker rather than assume full coverage.

Both posts are written as short warnings rather than case studies, ending on the same practical step — check the controls and the exclusions before the claim, not after.

Read original · Original 2

The stated goal is that the right consultant makes compliance less confusing, helps prioritize what matters and builds a clear path to meeting requirements. A companion post makes the related point that compliance vulnerabilities hide inside the technology stack when software and data protocols are not audited regularly.

The red-flags post is one of the few in the set written directly to a buyer choosing between providers, which makes it useful for understanding what Baker treats as good practice in her own field.

Read original · Original 2

Baker turns a reported banking concern into a concrete software-security test: can a company distinguish an AI agent from a person, and does the agent have its own limited, revocable access rather than a borrowed password? The post also notes a bank spokesperson’s contrary qualification, keeping the reported claims in perspective.

For brands planning a campaign

A B2B security campaign could use the agent-versus-human access question as a discussion prompt, then offer guidance on scoped agent credentials and monitoring.

Baker distinguishes a point-in-time view of control design (Type 1) from evidence that controls operated over time (Type 2). Her practical advice is to ask a major prospect which report it will accept before hiring an auditor.

For brands planning a campaign

A vCISO contributor could build a short buyer-readiness series around questions founders should settle before committing to an audit.

The post separates a pause in third-party assessments from continuing responsibilities, naming self-attestation, the SPRS score, annual affirmation, DFARS 7012, and NIST 800-171. It warns that an inflated score can carry False Claims Act risk.

For brands planning a campaign

For a defense-contractor audience, this could anchor an educational checklist on validating evidence before self-attestation.

Baker connects a ransomware advisory to a board-ready set of priorities—MFA, patching, segmentation, offline backups, and access reviews—then gives readers three questions to take to their IT provider, including when a restore was last tested.

For brands planning a campaign

A cybersecurity firm could adapt the question-led format for a practical board briefing or a provider-accountability campaign.

On SOC 2, the post draws the line between Type 1 (controls designed right, as of one date) and Type 2 (controls operating over a period, with the auditor's tests and results), and argues enterprise buyers want Type 2, so the practical move is to ask your biggest prospect which report they need before signing with an auditor. On CMMC, the post separates what a contract can require today from what a prime may ask for next: self-attestation is the current baseline, C3PAO certification continues even while the assessment phase is suspended, and a certificate lets a contractor show a prime proof rather than a promise.

Opens with the reader's question as the headline, gives the standard-setting body's definition, then states the decision rule.

For brands planning a campaign

Strong fit for content aimed at software vendors selling into enterprises and defense contractors weighing certification spend. The SOC 2 answer and the "ask your buyer which report" step are concrete enough to build a checklist or buyer-question asset around.

Examples: Source 1 · Source 2 · Source 3

The SafePay ransomware advisory becomes a five-item board slide (MFA on VPN, email and admin accounts; patching internet-facing devices; segmentation; offline backups with tested restores; access reviews) plus three questions for an IT provider.

• A lawsuit over a breached Defense Manpower Data Center file share becomes three lessons for any file share holding controlled unclassified information: encrypt it, log access and review the logs, and know your incident-reporting deadlines before an incident.
• An unconfirmed report of a personal AI agent connected to a secure defense network becomes three questions about approved tools, scoped agent access, and whether "an AI tool sent our data somewhere it shouldn't" sits in the incident response plan.

Sourcing is labeled. The defense-network post notes the claims are unconfirmed and denied; the data-center post labels the claims as allegations.

For brands planning a campaign

A repeatable news-jacking template for vendor campaigns: one event, one skeptical framing, three or four actions. Works especially well for partner content where a vendor can supply the checklist and the practitioner supplies the judgment.

Examples: Source 1 · Source 2 · Source 3

The position that recurs: an agent should never borrow a human's password — give it its own login, limit what it can do, and make it revocable in one step.

• From there it becomes an audit question software sellers should expect from enterprise buyers and SOC 2 auditors: can you tell agent traffic from human traffic?
• The same framing appears in the defense context, where an AI agent running on a contractor's admin credentials raises questions about which tools are approved for systems touching controlled unclassified information, whether agents get scoped access, and whether AI-related incidents are covered by the 72-hour reporting rule under DFARS 252.204-7012.
• An earlier post makes the general case that founders underestimate how AI audits affect operational security and data integrity.

For brands planning a campaign

A ready-made theme for identity, AI governance, and audit-readiness content. The "can you tell agent traffic from human traffic?" question is a usable hook for enterprise-facing campaigns, and the credential-scoping argument gives a vendor something concrete to co-sign.

Examples: Source 1 · Source 2 · Source 3

MFA, patching, segmentation, offline backups and access reviews are presented as what a board should be asked to fund, quoting Anthropic's deputy CISO calling for help with "the basics" rather than AI. On the compliance side, the same instinct produces a warning that self-attesting to CMMC Level 2 is a statement to the government, not a checkbox — the SPRS score, the annual affirmation, DFARS 7012 and NIST SP 800-171 still apply while C3PAO assessments are paused, and an inflated SPRS score is a False Claims Act risk.

For brands planning a campaign

Suits SMB and mid-market security campaigns where the buyer is weighing spend against basics. The SPRS-affirmation point is a sharper, higher-stakes angle for defense-industrial content.

Examples: Source 1 · Source 2

The first notes that a broker understands coverage but may not be able to determine whether security controls meet an underwriter's requirements, so MFA, the incident response plan and backup architecture should be verified before you need to rely on the policy. The second points to coverage gaps behind sub-limits, waiting periods and exclusions, naming social engineering fraud, business interruption and nation-state-related incidents as areas with meaningful limitations, and recommends reviewing the fine print with a broker before assuming full coverage.

For brands planning a campaign

Pairs naturally with MFA, incident-response and backup vendors, and with renewal-season timing. The "who checks the controls" question gives a security provider a clear role in a process the buyer already runs.

Examples: Source 1 · Source 2

The closing position is that the right consultant should make compliance less confusing, help prioritize gaps, and build a path to meeting requirements. A separate short post makes the related point that compliance vulnerabilities hide in the technology stack when software and data protocols are not audited regularly.

For brands planning a campaign

Useful for services firms and platforms that want to define what good looks like in a crowded market. The three red flags convert easily into an evaluation checklist a buyer can carry into a sales conversation.

Examples: Source 1 · Source 2

The recurring commercial call to action elsewhere in the corpus points the same direction: a free CMMC assessment link, or booking a call when compliance is stalling a deal.

For brands planning a campaign

Shows a low-friction lead-generation habit alongside the editorial posts. A vendor partnership could attach a co-branded self-assessment or scan as the campaign's first step.

Examples: Source 1

Posts carry calls to action for a free CMMC assessment and for booking a call, including the offer to "check [file shares holding CUI] against your SSP." The posts do not state her role at the firm, so whether this is ownership, employment, or another arrangement is unstated.

View example

She links to the tool. Whether she built, owns, or otherwise works with it is not stated.

View example

The posts do not state what her connection to it is.

View example

El Amrani's post on the Cyber Resilience Act walks through the 24-hour and 72-hour reporting clocks and asks who inside a company owns the decision before the clock starts, which is the same instinct Baker applies to AI agents and the DFARS 72-hour rule.

El Amrani's material centers on product security engineering and regulation timelines, with attention to sandboxing, telemetry and whether a company can even detect that a vulnerability is being exploited. Baker stays closer to U.S. compliance frameworks and the buyer-facing question of which report or certificate a customer or prime will accept.

Read original · Original 2 · Original 3 · Original 4

Pribanic's post on server misconfigurations 1 open ports and default passwords as the quiet cause of major breaches 1 sits in the same territory as Baker's argument that encryption, logging and tested backups are what protect a file share.

Pribanic works from IT hygiene and foundational risk assessment, and the offer is a hands-on assessment and control implementation. Baker works from the compliance framework outward, answering audit and certification questions and explaining the consequence that follows an inaccurate attestation.

Read original · Original 2 · Original 3 · Original 4

Levine's post on the security posture of third-party vendors makes the same shared-responsibility point Baker makes when she says a broker may not be able to judge whether your controls meet an underwriter's requirements.

Levine's posts read as high-level thought leadership and readiness messaging, including post-quantum solutions for the legal profession and an advisory board appointment. Baker's posts are more decision-focused, usually structured as a definition, a decision rule and a short set of checks.

Read original · Original 2 · Original 3 · Original 4

The reviewed posts work through the EU Cyber Resilience Act's 24-hour and 72-hour vulnerability and incident reporting clocks, and they focus on what has to happen inside a company before anyone can file: knowing a vulnerability is being actively exploited, agreeing a reporting threshold across security, engineering, product and legal, and naming who owns the decision.

• A second post points at agentic AI, sandboxing and harness engineering as the subject worth writing about.
• That fits a buyer whose audience builds products with digital elements in Europe, or who wants the incident-response-clock angle for a software vendor rather than a defense contractor.

The reviewed posts do not cover SOC 2 Type 1 versus Type 2, CMMC self-attestation, DFARS reporting, or cyber insurance sub-limits and exclusions, so a program on those topics would need to supply the substance and use this creator for the regulatory and engineering frame.

• The writing is also analytical rather than buyer-facing: it examines how a company is organised around product security, not which report a prospect will accept or whether a file share meets an SSP.
• You trade the founder-and-contractor decision framing and the short checklist-plus-single-call-to-action format for deeper regulatory specificity and a narrower technical audience.

Examples: Source 1 · Source 2

The reviewed posts make the case that an information security risk assessment is the foundational step toward a sustainable security program, name server misconfigurations such as open ports and default passwords as the quiet cause of major breaches, and offer directly to run the assessment and help implement the controls. That suits a campaign where the deliverable is a booked assessment conversation, with a light, plain-spoken tone that lowers the barrier for owners who are overwhelmed by breach headlines.

Depth per post is thinner.

• One post is a short invitation to reach out; the other links out to a blog rather than doing the analysis in the post.
• Neither reviewed post works through a compliance framework, so SOC 2 report choice, CMMC status, DFARS obligations and insurance exclusions would all have to come from the buyer.
• The tone is informal and promotional, which is an asset for SMB outreach and a liability if the campaign needs to read as neutral third-party judgment.
• You trade the evidence-led explainer format and framework coverage this brief depends on for approachability and a direct service offer.

Examples: Source 1 · Source 2

The reviewed posts tie third-party vendor risk to CISO collaboration, announce an advisory board appointment for a former legal-industry executive, and describe aligning quantum-proof solutions for law firms and enterprise customers through SonKsuru and QuSmart.AI, with a Forbes Technology Council affiliation. That fits a buyer whose brief is post-quantum readiness, third-party risk, or security go-to-market for professional-services firms, and who wants the credibility of a named company platform rather than an independent commentator.

The reviewed posts read as company and appointment announcements — welcoming an advisor, describing how solutions are being aligned — rather than buyer-facing explainers that answer a decision question.

• They date from 2023, so the framing is older than the current CMMC and AI-agent material this brief is built on, and they contain no SOC 2, CMMC, DFARS or cyber insurance content.
• You trade current, independent compliance judgment for an established vendor platform and a clear vertical focus.

Examples: Source 1 · Source 2