ZeeTwii

Life's Too Short for Frog DNA

Nick Ashworth — Fri, 18 Jul 2025 00:00:00 GMT

"Life's Too Short for Frog DNA": An Exclusive Interview with Rob "The Boss" Sivilli, CEO of Cost Plus Technologies

Posted to the Cost Plus Technologies Unoffical Blog | Cleared for Public Release: Yes, Obviously | Distribution: Unlimited (Except to InGen)

The following interview was conducted by Dr. James Nicholas Ashworth, D.Sc., Ph.D., Senior Distinguished Fellow of Computational Defense Epistemology, Emeritus Adjunct Chair of Applied Systems Irreversibility, and a man who has personally witnessed more PowerPoint briefings than any human being should reasonably survive. Dr. Ashworth sat down with Rob "The Boss" Sivilli, Founder, Chief Executive Officer, Chief Vision Officer, Chief Compliance Officer (self-certified), and Acting Director of Intersubjective Alignment at Cost Plus Technologies, to discuss the company's groundbreaking paleobiological defense portfolio, the competitive landscape, and why Mr. Sivilli has approximately zero respect for John Hammond.

Dr. James Nicholas Ashworth, D.Sc.: Rob, thank you for making time. I understand your calendar is, to use the technical term, absolutely catastrophic.

Rob "The Boss" Sivilli: Catastrophic is generous. Do you know how many contract modification requests I signed this morning? Neither do I — that's what the Director of Strategic Bureaucratic Optimization is for. But yes, I carved out forty-five minutes. I also billed it to a cost-plus line item, so technically the taxpayer is paying for this interview. You're welcome, America.

Dr. Ashworth: Let's begin at the beginning. Cost Plus Technologies has made a significant push into the dinosaur-based defense solutions market. Walk me through the origin of that strategic decision.

Sivilli: About three years ago, I was sitting in a Congressional hearing — as a witness, not a defendant, I want to be very clear about that — and a two-star general leans over to me and says, "Rob, what we really need is a heavy ground dominance platform that cannot be reasoned with, does not require a union contract, and will absolutely not unionize." And I said, "General, I have exactly one phone call to make."

That call launched the ALLOSAUR Heavy Ground Dominance Platform. Fourteen meters of validated lethality. Section 889 compliant. NDAA-friendly. Fully domestic — well, Cretaceous domestic, which our legal team assures us qualifies.

Dr. Ashworth: As someone with a background in computer science, I find myself curious about the systems architecture underlying the genomic pipeline. But I suspect you're going to redirect me to the DNA sourcing question first.

Sivilli: (leaning forward)

You suspect correctly. And James — may I call you James? — as a fellow PhD, I know you appreciate precision. So let me be precise.

Bird DNA. B-I-R-D.

We use bird DNA. Specifically, a proprietary avian-sourced genomic baseline that our science team — God bless them, every one — has spent years refining. The ALLOSAUR is, taxonomically speaking, a very large bird. This is not a joke. This is a phylogenetic fact that we have weaponized. And before you ask: yes, we documented the ontology. It's in the data dictionary. You can have a copy if you're cleared.

Now, I don't want to name names —

Dr. Ashworth: You're going to name names.

Sivilli: I'm going to name names. John Hammond. InGen Corporation. The man built a theme park — a theme park — on Isla Nublar using frog DNA. Frog. DNA. Do you know what frog DNA does to a genome sequence? It introduces sex-switching vulnerabilities into your population. Your entire dinosaur workforce becomes capable of spontaneous reproductive reorganization. That is not a feature. That is a catastrophic program management failure dressed up as "chaos theory."

We submitted a white paper on this to DARPA in 2019. They said it was "outside current solicitation parameters." InGen failed to win a single DoD contract. I'm just saying those two facts are related.

Dr. Ashworth: To be fair — and I want to be fair, it's a professional obligation — InGen was primarily a commercial venture.

Sivilli: (long pause)

A commercial venture. On a classified island. With classified animals. Funded by — and I looked into this — a combination of private investors and what my attorneys describe as "aggressively ambiguous offshore entities." Meanwhile, we filed proper CAGE codes, registered in SAM.gov, and completed our CMMC Level 2 assessment before we sequenced a single strand of theropod genome. That's the difference between a professional defense contractor and a billionaire with a god complex and a helicopter.

Also, his park had a 100% guest casualty rate on opening weekend. Our ALLOSAUR units have a controlled casualty rate. It's in the spec.

Dr. Ashworth: I've spent enough time in acquisition to know that "it's in the spec" is doing enormous load-bearing work in that sentence. How do you respond to critics who suggest a 14-meter carnivorous theropod is a non-trivial liability risk?

Sivilli: I respond by referring them to our Legal Disclaimers page, which I think speaks for itself. We have indemnification language in that document that covers acts of God, acts of dinosaur, and at least three categories of event that don't have legal names yet.

More substantively: the ALLOSAUR is a managed system. It operates within defined engagement parameters. It has been behaviorally conditioned using a proprietary positive-reinforcement protocol that I cannot discuss in detail because it is covered under a CRADA with a major land-grant university and also because it involves a truly alarming amount of raw beef.

InGen tried to manage their assets with electric fences. We use a combination of operant conditioning, geofencing, and what our Chief Behavioral Systems Architect calls "strategic caloric leverage." It's more humane, more effective, and — critically — it does not fail when someone cuts the power.

Dr. Ashworth: Hammond has said, on the record, that he "spared no expense." You've described this characterization as, and I'm quoting your internal Slack here, "delusional."

Sivilli: (audible exhale)

He spared every expense. The man had a park full of apex predators and his entire security architecture ran on a single workstation operated by a disgruntled employee with a Barbasol can. His IT security posture would not pass a basic RMF assessment. It would not pass a conversation about a basic RMF assessment. James, you and I have both sat through ATO reviews. You know that what Hammond built would have been dead on arrival at the first ISSO meeting.

We operate on a zero-trust architecture. Zero trust for the network. Zero trust for the endpoints. Zero trust for the dinosaurs, frankly — that's just good program management. Our ALLOSAUR units do not have access to the corporate VLAN. That is a design decision I am proud of.

Dr. Ashworth: Speaking as someone who has watched DoD IT infrastructure decisions get made in real time — that is the correct decision. Now, what else is in the pipeline?

Sivilli: I can speak obliquely and let the procurement community read between the lines, which is how we've always done business — as you well know.

We have an upcoming platform in the aerial dominance space. Avian genomics again — bird DNA, not frog, I cannot stress this enough — that addresses a capability gap in the low-altitude surveillance and interdiction domain. I will not confirm or deny whether it has feathers. I will confirm that it is Section 508 compliant, which I think demonstrates our commitment to inclusive design across all biological platforms.

We are also exploring aquatic options. I won't say more than that. I will say that InGen had the IP rights to a certain marine-based platform and let them lapse. Their loss. Our CAGE code's gain.

Dr. Ashworth: Final question, and I'll make it appropriately grandiose given the setting. What is the core message you want the defense acquisition community to take away from Cost Plus Technologies?

Sivilli: We are the professionals. We file our paperwork. We use the correct DNA. We do not build theme parks and call them research facilities. We do not rely on a single power grid to contain assets that can bite through a Ford Explorer.

When you issue that next BAA — when you're staring at a capability gap that no conventional system can fill — we want you to think of us. We want you to think: "There's a company out there that took the Cretaceous seriously, that did the genomics right, that built a cost-plus contract structure so elegant it should be in a museum."

And then we want you to call our Business Development team, who will be delighted to take your call between the hours of 10 AM and 3 PM, Monday through Thursday, excluding federal holidays and any day following a scheduled ALLOSAUR feeding.

Dr. Ashworth: Rob, on behalf of the defense acquisition community, and as a fellow survivor of the federal IT enterprise, thank you.

Sivilli: Charge it to the program.

Cost Plus Technologies is a registered defense contractor. The ALLOSAUR Heavy Ground Dominance Platform is currently in Limited User Testing. John Hammond was unavailable for comment. InGen Corporation did not respond to a request for comment, possibly because they no longer exist as a functioning entity. Dr. Ashworth's views are his own and do not represent the official position of any agency, department, command, or sentient apex predator. This interview has been edited for length, clarity, and OPSEC review.

The Frontal Lobe Is Not a Load-Bearing Structure

Nick Ashworth — Sat, 10 Jan 2026 00:00:00 GMT

"The Frontal Lobe Is Not a Load-Bearing Structure": An Exclusive Interview with Rob "The Boss" Sivilli on the PTERODACTYL Incident

Posted to the Cost Plus Technologies Unoffical Blog | Cleared for Public Release: After Significant Legal Review | Distribution: Unlimited (Except to the JAG Office, Pending)

Six months after our first interview, Cost Plus Technologies finds itself at the center of a controversy that has made the rounds in defense acquisition circles, three naval aviator group chats, and at least one Reddit thread that has since been removed for reasons the company declines to specify. Dr. James Nicholas Ashworth, D.Sc., returned to speak with Rob "The Boss" Sivilli about the deployment of the PTERODACTYL Counter-Unmanned Aerial System to CVN-87, the subsequent incident now being referred to in certain circles as "The Bite of '87," and what the company's path forward looks like. Mr. Sivilli requested that we note, for the record, that he is "completely calm" about all of this.

Dr. James Nicholas Ashworth, D.Sc.: Rob. CVN-87. Walk me through it.

Rob "The Boss" Sivilli: I want to start by saying that Petty Officer Fischbach is doing extremely well. He is responsive, he is communicative, and according to the ship's medical officer, he is living proof that the frontal lobe is, in the words of the official medical report, "not a strictly load-bearing neurological structure." Those are the doctor's words. I'm just repeating them.

Dr. Ashworth: The sailor who was bitten.

Sivilli: Who was grazed. The after-action report uses the word "grazed." We fought very hard for that word.

Dr. Ashworth: The official NCIS report uses "bite."

Sivilli: The official NCIS report uses a lot of words. It also uses "unauthorized engagement envelope exceedance," which I think more accurately captures what happened from a systems perspective. The PTERODACTYL unit was operating within its defined behavioral parameters — we think — and Petty Officer Fischbach made a movement near the flight deck that the unit's threat-discrimination algorithm interpreted as, and I'm going to use the technical term here, a drone.

Dr. Ashworth: He was carrying a coffee mug.

Sivilli: A very fast coffee mug. We are reviewing the sensor calibration.

Dr. Ashworth: Let's back up. For readers who aren't familiar, describe the PTERODACTYL system.

Sivilli: Absolutely. The PTERODACTYL Precision Threat Elimination and Rapid Organic Denial of Aerial Craft Technology with Ychromosomal Logic — the acronym took three weeks and I stand behind it — is Cost Plus Technologies' answer to the low-altitude UAS threat. It is a pterosaur-derived, avian-genomics-based counter-drone platform with a wingspan of approximately four meters, a top airspeed that we are contractually prohibited from disclosing, and a threat-interdiction methodology that I would describe as "extremely thorough."

It uses bird DNA. I cannot stress this enough. We are not InGen. We did not cut corners with amphibian genomics. Every PTERODACTYL unit was sequenced from a certified avian baseline, which means it is, taxonomically speaking, a very large, very fast, very motivated seabird with a beak the length of a regulation softball bat.

Dr. Ashworth: And it was deployed to CVN-87 specifically.

Sivilli: CVN-87, yes. The USS Gerald R. Ford-class vessel currently under the command of Captain Jeremy Fitzgerald, who I want to say is an outstanding naval officer who has handled this entire situation with extraordinary professionalism, especially given his, let's call it, personal history with bite-related incidents. The man is a survivor in every sense of the word. We have enormous respect for him.

Dr. Ashworth: I'm not going to ask about that.

Sivilli: Appreciate it.

Dr. Ashworth: The ship's designation — CVN-87. You didn't flag that as potentially inauspicious before deployment?

Sivilli: (long pause)

Our Program Manager flagged it. In writing. There is a memo. I chose not to act on the memo, which in hindsight I acknowledge was a program management decision that did not age well. It is now a standing agenda item in our weekly risk review. We have a new policy: any contract involving the number 87 in the hull designation, contract vehicle number, or CLIN structure triggers an automatic safety review. We are calling it the Fitzgerald Protocol.

Dr. Ashworth: How is Captain Fitzgerald responding to having a safety protocol named after him?

Sivilli: He has not responded to our emails. His XO sent us a very terse one-sentence reply that I am advised not to quote in a public forum.

Dr. Ashworth: Let's talk about Petty Officer Fischbach. How is he actually doing?

Sivilli: Mark is doing great. Genuinely. He was remarkably composed during the incident — by all accounts he stood there for approximately twenty seconds with a completely blank expression, just processing what had happened, and then looked directly at the ship's camera system and said, and I am told this is an exact quote: "Was that the Bite of '87?"

Which, honestly, is a level of presence of mind under pressure that I find impressive. That man has a future in crisis communications if he ever leaves the Navy.

We have also, at the company's expense, sent him a very nice fruit basket and a Kevlar flight deck helmet with enhanced cranial protection. The helmet is rated for pterosaur beak impact up to approximately forty miles per hour. The PTERODACTYL unit in question was traveling at approximately forty-three miles per hour at the moment of contact. We are working on the next revision.

Dr. Ashworth: I feel obligated to ask: is the frontal lobe comment from the medical officer a real assessment, or is that some kind of legal positioning?

Sivilli: It's real, it's in the report, and it is also — and I say this with tremendous gratitude — almost verbatim language from a DoD medical document that was used in a completely unrelated incident in 1987. I have cited it in our incident response. Our legal team called it "an inspired use of precedent."

Dr. Ashworth: InGen must be having a field day.

Sivilli: (visibly tightening)

InGen. Yes. We have been made aware that a representative of InGen's successor entity — I will not name them, they know who they are — sent a statement to two defense trade publications suggesting that this incident, quote, "validates concerns about the use of unproven biological platforms in active fleet environments."

This from a company whose flagship product ate a lawyer on its opening day. On a toilet. Their entire liability record is one long argument against their own credibility, and they have the audacity to issue statements about our quality assurance process. Their assets used frog DNA. Their assets could spontaneously reorganize their own reproductive biology. Our asset had a sensor calibration issue. These are not equivalent problems.

Dr. Ashworth: To be scrupulously fair: their assets did not bite a sailor on a carrier.

Sivilli: Their assets ate guests. Paying guests. On a commercial property open to the public. We had one, single, documented beak-to-cranium contact event on a hardened military vessel in a controlled operational environment, and the sailor survived with full cognitive function. I'll take that trade every time over whatever liability exposure comes with having a Velociraptor loose in a visitor center.

Also — and I want this in the record — the PTERODACTYL successfully intercepted fourteen Group 2 UAS threats in the three days prior to the Fischbach incident. Fourteen. It has a 100% interdiction rate against actual targets. The Fischbach event represents a 6.7% friendly-contact rate against non-drone targets on the flight deck. In acquisition terms, that is well within acceptable system immaturity thresholds for a Milestone B program.

Dr. Ashworth: I'm confident no one has ever used those words in that order before in a Senate hearing.

Sivilli: We have a hearing in six weeks. I have already used them in that order, in a draft opening statement, twice.

Dr. Ashworth: Any modifications to the program going forward?

Sivilli: Several. First, the threat-discrimination logic is being updated to deprioritize slow-moving, low-thermal-signature objects below a certain altitude threshold — we're calling this the "coffee mug exclusion zone." Second, all flight deck personnel will be issued the aforementioned cranial protection. Third, we are looking at a behavioral conditioning update to better distinguish between a DJI Phantom 4 and a human being holding a beverage.

We are also — and this is not officially announced — exploring a daytime restriction protocol for the PTERODACTYL units while the ship is at general quarters. We will limit their free-roaming mode to nighttime operations only, to prevent any further engagement envelope ambiguity. Our Chief Behavioral Systems Architect calls this the "servo lockup prevention protocol." We call it "not letting the pterosaur walk around the carrier in the daytime." It amounts to the same thing.

Dr. Ashworth: That is essentially the exact lesson that Freddy Fazbear's Pizza learned in 1987.

Sivilli: (long silence)

Yes. It is. And I want to be clear: we arrived at this independently, through rigorous after-action analysis, and not because a DoD IG investigator left a printout of a Wikipedia article on my desk with the relevant paragraph highlighted in yellow. That did not happen.

Dr. Ashworth: Final question. What do you want the defense acquisition community to take away from this?

Sivilli: That biological platforms in operationally contested environments require iterative refinement, and that iterative refinement sometimes involves a sailor named Mark and a very enthusiastic pterosaur. That is not a failure. That is the developmental process. Every transformative defense capability in history has had a Fischbach moment. We have learned from ours. We have documented our lessons learned, updated our TTP's, and submitted a contract modification request for the enhanced helmet program that I am confident will be approved, funded, and delivered before CVN-87 completes its current deployment.

We are the professionals. We use bird DNA. We file our paperwork. And when our system bites a sailor, we send fruit.

Dr. Ashworth: Rob, thank you. And please pass along our best wishes to Petty Officer Fischbach.

Sivilli: He knows. He's been following the blog.

Cost Plus Technologies is a registered defense contractor. The PTERODACTYL CUAS platform is currently in a modified Limited User Testing phase pending sensor recalibration. Petty Officer Mark Fischbach has been medically cleared for full duty and has reportedly requested a transfer to a non-pterosaur-equipped vessel, which is under review. Captain Jeremy Fitzgerald declined to comment. The CVN-87 Fitzgerald Protocol has been submitted for formal inclusion in the Cost Plus Technologies Quality Management System. InGen's successor entity did not respond to a request for comment, which is statistically consistent with their pattern of behavior. Dr. Ashworth's views are his own. This interview has been edited for length, clarity, and ongoing NCIS coordination.

So you want to steal a model?

Nick Ashworth — Fri, 27 Feb 2026 00:00:00 GMT

So you want to steal a model?

Early this week, Anthropic released a report about the latest distillation attack they detected against there models. While they don't explictly go over the cost of performing the attacks, it does bring to light a glaring error with how we currently think of both AI policy in the US, and what it means for LLM vendors as an industry.

Model Distillation

Before going further, let's define what we mean by distillation in this space. Model distillation is the process by which you train a smaller model on the outputs of a larger model. Do this repeatedly over a long enough time period, and eventually the smaller model will perform roughly as well as the larger model, without increasing in size.

Where this transitions from a technique to an attack in Antrhopic's case is that both they, OpenAI, Google, and XAI, all keep their models private. When you buy access to their models, their API, and most of their revenue stream is designed around keeping said model private and unable to be easily duplicated or recreated. By asking a wide varity of questions that stimulate a wide variety of responses, you can train a smaller model on the outputs of the larger model, and eventually have a model that performs just as well as the larger model, without needing direct access to the larger model. This is why Anthropic is calling it an attack, because it is a direct attack on their business model, and the way they make money. If you can steal their model, then you don't need to pay them for access to it anymore.

Distillation attacks do raise a weird truth about the current state of the LLM industry: so much of the value of these companies is in the data they have access to, and the models they have trained on that data. If you can easily recreate the model, is there really any justification for the high evaulations many of these companies have? All of our current policies and regulations are based on the idea that the models these companies have are unique and can't be easily replicated. If that is not the case, then we need to rethink how we value these companies, and how we regulate them.

The Cost of Distillation

While Anthropic doesn't go into the cost of performing the attack, we can make some educated guesses based on the information they do provide. While they don't list the exact interactions, they do provide how many interactions they flagged as being part of the attack, and how many interactions they had in total. Based on this, we can estimate the cost of performing the attack, and how much it would cost to steal a model like Claude.

Cost Formula

Anthropic's API is priced based on the number of tokens you use. So to try and estimate the cost, we need to figure out how many tokens were used in the attack. However, Anthropic only provides the number of interactions, not the number of tokens. To get around this, we have to estimate the number of tokens per interaction. A quick google search shows the average number of tokens per conversation can vary widely, with 4,000 to 8,000 tokens being a common range. Given that Anthropic lists over 24,000 accounts possibly being involved, we can assume the attackers were trying to make all the interactions look like normal conversations. To make the math easier, we'll assume every interaction uses the high range of the estimate, 8,000 tokens. This is probaly an overestimate, but it should give us a good upper bound on the cost of performing the attack. With that in mind, we can use the following formula to estimate the cost of performing the attack:

Cost = (Number of Interactions) * (Tokens per Interaction) * (Cost per Token)

Where:

Cost per Token: We'll use Anthropic's most expensive model, Opus 4.6, which has a cost of $25 per million tokens, or $0.000025 per token.
Tokens per Interaction: As mentioned above, we'll use 8,000 tokens per interaction.
Number of Interactions: Anthropic lists 16 million interactions in total that were flagged as part of the attack.

Using these numbers, we get a total cost of the attack to be: $3,200,000. $3.2 million dollars is not a lot of money for a company or nation state to spend on stealing a model that could potentially be worth billions of dollars. This is especially true when you consider that the cost of training a model like Claude from scratch would be in the hundreds of millions of dollars, if not more.

In a weird way, this is actually a problem for Anthropic and other cloud based model vendors. Even if they double the price of their API, the cost of performing the attack would still be so cheap that it would still be worth it for a company or nation state to steal the model. In a weird way, the attacker is actually the best customer they will every have because there is no price they won't pay for API access. This means they can't use the classic Steam approch of making it easier to buy the game than to steal it, because there is no price they can set that would discourage stealing the model and not also cause them to lose customers.

Indidual Breakdown

Antrhopic breaks down the total attack and lists several companies by name, so I thought it would be interesting to break down the cost of the attack for each company using the same formula as above. This is not an exact science, as we don't know individual token usage or how many tokens were input vs output, but it should give us a good estimate of the cost of the attack for each company. Here is the breakdown:

Company Name	Number of Interactions	Estimated Cost of Attack
DeepSeek	150,000	$30,000
Moonshot AI	3,400,000	$680,000
MiniMax	13,000,000	$2,600,000

Conclusion

The cost of performing a distillation attack against a model like Claude is shockingly low, especially when you consider the potential value of the model. This raises some serious questions about the current state of the LLM industry, and how we value these companies. If the models these companies have are not unique and can be easily replicated, then we need to rethink how we regulate them, and how we value them as companies. Model vendors need to start thinking about how they can detect and prevent these types of attacks, which will require more heavily analysis and monitoring of their API usage, response patterns, and user behavior. This is not an easy problem to solve, but it is one that needs to be addressed if these companies want to continue to be successful in the long term.

So you think you saw a drone?

Nick Ashworth — Tue, 31 Mar 2026 00:00:00 GMT

So you think you saw a drone?

A couple weeks ago, we started getting more news articles claiming drone sightings within CONUS (Continental United States) military bases. This has spawned a lot of panic and speculation about what these drones could be, who is flying them, and what their intentions are. However, most of these sightings are likely misidentified objects rather than actual drones. I've done incident response to drone sightings for years, and I can tell you that the vast majority of these reports turn out to be false alarms. In this post, I'll explain why most drone sightings are misidentifications and what you can do to avoid falling for the same trap.

It has lights, how hard can this be?

People tend to assume that if they see something with lights in the sky, it must be a drone. However, especially in US airspace, there are many objects that can have lights and be mistaken for drones. For example, airplanes, helicopters, weather balloons, satellites, and even stars all either have lights or illuminate and have all been mistaken for drones at some point. People don't normally look up at the sky, and so when they do, they will often misinterpret what they see. Figuring out how far away and how large an object is can be surprisingly difficult, especially at night. A small drone flying close to you can look very similar to a large airplane flying far away. This even extends to movements, people will often mistake a aircraft for drones because they misinterpret the movements of the object and assume it is hovering or moving too slowly to be anything else.

There is a classic tell for these kinds of misidentifications, which is that the meme of standing there menacingly. In the case reported by ABC news, the witnesses are claiming that a swarm of drones have repeatedly violated US military airspace for the sole purpose of hovering with their lights on while base security panics. This is not how actual drone attacks work, and not something any threat actor would do. Drones are generally used for either surveillance or attacks, and in either case, they would not be hovering in place with their lights on for extended periods of time. No malicious actor would want to draw attention to themselves by hovering in place with their lights on, especially in a high-security area like a military base.

One funny story I have from my own experience is that I once had to rush to respond to a drone incident in the NCR (National Capital Region) because someone reported that a drone was dropping unknown "payloads" on the roof of a instillation. We set up multiple cameras and RF scanners to try and catch the drone in the act, but all we ended up finding was a bird that was flying around and dropping clams on the roof to break their shells. That was the mysterious "payload" that was being reported.