The meetup slide advertised CSS26-MASSCC for free admission by October 8. The code is publicly documented on MassCyberCenter's events page, and MASSCC is MassCyberCenter itself, a division of the Massachusetts Technology Collaborative and a summit partner. CSS26-<PARTNER> is the organizer's standard partner-comp pattern; other live instances include CSS26-OS and CSS26-WICT.

Verified on 2026-09-17 by registering: the code produces a Standard Attendee order at 0.00 USD, payment status "No Fees", registration Confirmed. Eligibility screening passed without challenge.

The verification is dated and does not generalise forward. It was used 21 days before the advertised "by Oct. 8" deadline, and a partner comp is the kind of thing that runs to a quiet cap rather than to its stated date. Anyone trying it in October should expect to pay and be pleasantly surprised, not the reverse.

Until registering, this was unverifiable from outside, because the page renders identically for the real code and for a deliberately invalid control code; validation happens only after the form is submitted, which is either an absence of client-side validation or a deliberate refusal to confirm which partner codes exist. The published offer terms are non-transferable, new registrations only, no retroactive use, subject to eligibility review.

The eligibility terms are worth reading after registering rather than before, because they apply afterwards. Admission is restricted to active practitioners responsible for defending an enterprise; students, interns, educators, anyone not currently employed in IT, and anyone in a sales or marketing role are excluded. All registrations are subject to review, and the terms add that if the organizers cannot verify an identity from what was given at registration, attendance may be cancelled.

The sentence that matters for a comp registration: "these qualifications pertain to all attendees, including members of our partner organizations." A partner code buys the fee, not the seat. A confirmed registration is therefore a claim on a seat pending review, not an admission ticket, and the review happens on their schedule rather than at the door.

The September note recorded slot counts, three vendor sessions with times, the featured speakers, and the panels' thin rosters. It did not record the full running order, so the drift below is what can be read against what was written down, not a full slot-by-slot diff.

• The three tracked sessions all moved earlier. Sonatype 10:10 → 09:40, Mimecast 12:15 → 11:45, TrendAI 14:15 → 12:05. TrendAI moved from the afternoon to before lunch; its old 14:15 slot is now Securiti AI.
• The day grew from about fourteen content slots to nineteen. Vendor slots went from nine to thirteen, plus the STEMatch presentation, which is a founder presenting his own company. The two-thirds figure in the description holds (13 of 19 is 68%).
• Vendors the snapshot did not record. Specops (Outpost24), Gambit Security, Huntress, Sublime Security, Expel, Securiti AI, Fidelis Security, Netskope, QuellSecure and Nox Security. In September four slots were bare vendor names; which of these ten were among them was not written down.
• A featured speaker vanished. Randy Rose (Center for Internet Security) was one of three featured speakers in September and holds no slot in the final agenda. Tetreault (CISA) and Diamond (CyberSN) remain.
• Panels filled in, unevenly. The AI panel now names a moderator and three panelists. The supply-chain panel still names two people, neither marked as moderator. The third panel turned out to be about team diversity.
• The Securiti speaker is listed under Veeam. Todd Vancil is titled VP Sales Engineering and Enablement at Securiti, with the organisation field reading Veeam. That plausibly reflects Veeam's announced acquisition of Securiti AI; not checked here.
• Time zone. The page labels the day EST. In October Boston is on EDT. Left as published in the table above.

Mythos-Ready: Building a Security Program for the AI Vulnerability Storm. Irina Tishelman, Principal Sales Engineer. Listed at 10:10 in September.

AI-accelerated vulnerability discovery and supply-chain component gating. Worth recording that the abstract is pegged to a model release, opening with the claim that the arrival of Anthropic Mythos "fundamentally altered the cyber threat landscape." That is the vendor's framing, not a finding, and the pegging of a security posture to a named model version is itself the thing to watch.

The final abstract adds numbers, all the vendor's: vulnerability volume predicted to rise two to ten times, and NVD said to be prioritising only KEV entries. The prescription is to move control to component selection at the start of the supply chain rather than to remediation downstream. The LiteLLM PyPI compromise in Maintaining Agent Sandbox Systems is the case for that, and also its test: the malicious wheels were new releases of a trusted package, published with stolen maintainer credentials and live for about three hours. A gate that admits a package by name admits them; one that pins by hash does not.

Verifying Identity in the Age of AI-Powered Impersonation. Ryan Burkhard, Specops Software (Outpost24). Cloned voice and deepfake video aimed at new hires and the help desk; the gap named is verification and account recovery. The only session on the deepfake item in the organizers' AI theme, and its subject is human identity; agents do not appear.

Edwin Moreno, Mimecast. Listed at 12:15 in September.

The closest thing on the agenda to agent identity and authorization. Its abstract is the argument in one line: agents now hold credentials, reach systems, and decide on behalf of the business, so apply one measurement approach to human and agent behaviour alike.

Whether a single measure over both is right is exactly the question trust boundaries as a typed category pushes against: a delegated agent identity and a human identity differ in what may be inferred when each misbehaves.

The final abstract names the measure's three parts, visibility, policy adherence and detection, and argues for residual risk over benchmark scores. Visibility of an agent presupposes knowing whom it acts for, the question Agent Identity and Attestation treats.

From Vibe Coding to Wire Fraud: An AI Security Blueprint for the Governance Gap. Phil Atkinson, TrendAI. Listed at 14:15 in September; now before lunch.

The most technically specific slot on the agenda. Names indirect prompt injection, system prompt leakage, and Model Context Protocol abuse directly, and proposes a control plane enforcing before, during and after the model runs. Adjacent to elenctic vibe code review on the authorship side and to the egress-proxy thread on the enforcement side.

The final abstract adds the phrase "owned blast radius". Indirect prompt injection is the case worked through in Tainted Data in LLM Pipelines, whose conclusion is that enforcement has to leave the model. MCP abuse lands on the authorization gap in Agent Permission Guardrails.

Security & Networking ReAImagined. Mike Brockney, Netskope. Ten minutes on an AI Gateway, an Agentic Broker exposed as an MCP server, and blocking of prompt injection and jailbreaks. A broker in front of MCP is a policy decision point by another name; the question to ask is whether it decides per tool call or per session, the distinction Agent Permission Guardrails draws.

Agents Don't Belong on the Endpoint. Aram Peles Chen, CEO. Move agents off human endpoints into isolated cloud environments whose network, secrets, MCP servers, skills and packages are defined by policy. This is the bulkhead in Sandboxing AI Coding Agents with FreeBSD Jails sold as a service. The claim to test is secrets: Agent Sandbox Architectures finds secret custody the axis the field has not built.

Securing the next 250: Cybersecurity Practices for All Organizations. Mike Tetreault, Cybersecurity Advisor, CISA.

Dan Daggett, Gambit Security. Backups that would not survive a real attacker; resilience framed as shared goals in the DevOps manner.

Moderator William H. Guenther (ACSC); Samuel Guyer (Veracode), Jonathan DiVincenzo (Impart Security), Shashi Mysore (Zscaler). All three panelists work for vendors, so the panel is not the non-vendor counterweight the September count assumed.

Chris Straubs, Sublime Security. An organisation-specific detection layer for email, with Snowflake and Elastic named as integrations.

Supply Chains, Vendors, and Identities. Thomas Hart (ISACA board), John Petrozzelli (MassCyberCenter board). Two names; no moderator listed.

Todd Vancil, Securiti (listed organisation: Veeam). Cites Gartner's prediction that 60% of AI projects without AI-ready data are abandoned through 2026, and the 5% "future-built" figure from BCG.

Diversity in security teams. Moderator Deidre Diamond (CyberSN); Oyefunke Fayoyin (CISO, Devo and Strike48), Kelly Haydu (Vorlon).

Accessing an untapped, job-ready, cost-effective cybersecurity talent pool. Chris Zannetos, founder, STEMatch.

Anson Dorsey, CTO, QuellSecure. Contain encryption while it runs rather than recover afterwards; the inverse of the Gambit slot's emphasis.

Doug Domin, Supervisory Special Agent, FBI.