AI
“Bank of America hired 1,000 outside contractors to fix software vulnerabilities exposed by AI models including Anthropic's Mythos." (plus the post's supporting bullets: 3,000 existing staff; "more than $200 million" in Q3 on software fixes and AML compliance combined; 130 internal AI projects; $700 million in savings in 2027)”
Plain restatementBank of America engaged approximately 1,000 external contractors to patch software vulnerabilities surfaced by advanced AI models, one of which is Anthropic's Mythos. Those contractors work alongside roughly 3,000 existing internal staff. The bank's CEO stated third-quarter costs included a few hundred million dollars covering software remediation and anti-money-laundering compliance work together. The bank expects roughly 130 internal AI projects to deliver about $700 million in annual benefit in 2027.
This one largely checks out. Bank of America's CEO Brian Moynihan said on the record on October 14 2026, both on the bank's earnings call and in a CNBC interview, that the bank brought in about 1,000 outside contractors to patch software flaws exposed by advanced AI models including Anthropic's Mythos, working alongside about 3,000 existing staff. The bank's own published transcript also confirms the roughly 130 AI projects and the expectation of about $700 million in benefits next year. Three things are worth knowing that the post leaves out. The contractors were actually taken on in the second quarter of 2026 and are still there, not hired this month. The "more than $200 million" is the post's rendering of Moynihan's spoken "a couple hundred million plus," and he never split how much was AI patching versus anti-money-laundering work. Most significantly, in the same quarter the bank separately booked a $1.2 billion charge for anti-money-laundering penalties from US regulators, which the post does not mention. Nobody has published how many flaws Mythos actually found in the bank's code or how serious they were.
Bank of America [drifted from the evidence:] hired 1,000 [drifted from the evidence:] outside contractors to [drifted from the evidence:] fix software vulnerabilities [drifted from the evidence:] exposed by AI models [drifted from the evidence:] including Anthropic's Mythos." [drifted from the evidence:] (plus the post's supporting bullets: 3,000 existing staff; [drifted from the evidence:] "more than $200 million" [drifted from the evidence:] in Q3 on software [drifted from the evidence:] fixes and [drifted from the evidence:] AML compliance [drifted from the evidence:] combined; 130 internal AI projects; $700 million in [drifted from the evidence:] savings in 2027)
Bank of America [added by the neutral restatement:] engaged approximately 1,000 [added by the neutral restatement:] external contractors to [added by the neutral restatement:] patch software vulnerabilities [added by the neutral restatement:] surfaced by [added by the neutral restatement:] advanced AI models, [added by the neutral restatement:] one of which is Anthropic's Mythos. [added by the neutral restatement:] Those contractors work alongside roughly 3,000 existing [added by the neutral restatement:] internal staff. [added by the neutral restatement:] The bank's CEO stated third-quarter costs included a few hundred million [added by the neutral restatement:] dollars covering software [added by the neutral restatement:] remediation and [added by the neutral restatement:] anti-money-laundering compliance [added by the neutral restatement:] work together. The bank expects roughly 130 internal AI projects [added by the neutral restatement:] to deliver about $700 million in [added by the neutral restatement:] annual benefit in 2027.
Red-tinted words in the claim drifted from the evidence. Green-tinted words are what a neutral restatement needs.
The trace / claim to source
- Bank of America did engage about 1,000 outside contractors for software patching work. This comes from the CEO's own words on October 14 2026, not from an anonymous source or a leak.
- The 3,000 figure is accurate as stated. Moynihan described the contractors as working "on top of the 3,000 people we have in that area," meaning staff already assigned to that work, not the bank's whole security organisation.
- Anthropic's Mythos is real, and the post names it correctly. Anthropic itself announced the model in April 2026 under a restricted-access programme called Project Glasswing and said it did not plan to make the preview model generally available.
- The caption's phrasing "AI models including Anthropic's Mythos" correctly preserves what Moynihan said. He attributed the patching surge to frontier models generally and named Mythos as one of them, rather than to Mythos alone.
- The $700 million figure is accurate and appears in the bank's own transcript, where Moynihan projected a rise from roughly $250 to $300 million this year to $700 million next year across roughly 130 implemented AI projects.
- The post's characterisation of the Q3 cost as covering software fixes and anti-money-laundering compliance "combined" is correct. Moynihan gave one blended figure and did not split it.
- The post says the bank "hired" the contractors without a date. Moynihan specified they were put on in the second quarter of 2026 and remain today. Undated, the line reads as a fresh October hiring event when it describes a staffing decision made months earlier that is now being disclosed.
- "more than $200 million" renders Moynihan's spoken "a couple hundred million plus," which is an approximation offered verbally and not a reported accounting line. The substitution is numerically defensible as a floor, but it converts an imprecise estimate into a figure that reads as a disclosed number. The underlying spoken figure could reasonably mean a sum well above $200 million.
- The post presents the "more than $200 million" as the quarter's anti-money-laundering cost picture. In the same quarter the bank accrued a separate $1.2 billion charge for civil money penalties under consent orders with the OCC and FinCEN, which its own SEC filing records. A reader of the post alone would substantially understate the quarter's AML-related financial hit.
- Omitted qualifier (in the image text only, not the caption): the on-image line reads "software flaws exposed by Anthropic's AI models," dropping the word "including." That narrows a statement about frontier models in general, with Mythos named as one, into an attribution to Anthropic alone. The caption does not have this problem.
- Framing note that fits none of the standard names: the construction "Despite these costs, Bank of America expects its 130 internal AI projects to generate $700 million in savings in 2027" invites the reading that the AI savings offset the remediation bill. These are separate items in Moynihan's remarks. The savings figure covers a portfolio of internal efficiency projects, while the remediation spend is a security and compliance cost. Nothing in the transcript nets one against the other.
- How much of the "couple hundred million plus" was Mythos-related patching and how much was anti-money-laundering compliance work. Moynihan explicitly did not separate them, so the AI-attributable cost is not established by any source found.
- Whether the roughly 1,000 contractors were engaged solely for AI-surfaced vulnerabilities or also absorbed other remediation work. The statements group the effort broadly.
- The number, severity and nature of the vulnerabilities Mythos actually surfaced in Bank of America's proprietary and vendor code. No vulnerability count, no severity breakdown and no independent verification of the findings was published by either Bank of America or Anthropic.
- Whether the $700 million 2027 figure is a gross benefit or a net saving after the cost of running the AI projects. The transcript line refers to benefits from projects in the pipeline and separately to 2x to 3x returns over expense, which leaves the accounting boundary unclear.
- Which specific Mythos version or versions Bank of America had access to, and under what terms. Moynihan did not say, and the family includes a preview release and later versions.
The claim traces to on-the-record statements by Bank of America's chairman and CEO Brian Moynihan on October 14 2026, made both on the bank's third-quarter earnings call and in a subsequent CNBC television interview. CNBC reports that Bank of America hired an army of external contractors to help it patch software vulnerabilities exposed by advanced artificial intelligence models including Anthropic's Mythos, with Moynihan saying the company rushed earlier in the year to bolster a wide spectrum of its internal and vendor code using thousands of employees and outside personnel. His direct words were that the bank put on 1,000 contractors to do the patching work in the second quarter, that those people remain today, and that they sit on top of the 3,000 people the bank already has in that area. On the cost figure, The Register reports Moynihan said during the Q3 earnings call that the bank spent "a couple of hundred million plus" in remediation costs related to Mythos and to the anti-money laundering orders issued that week, and that Moynihan did not break down how much of that figure was attributable to each. On the AI savings figure, Bank of America's own earnings call transcript records Moynihan saying the bank expects that number to go from the $250, $300 this year to $700 million next year in terms of projects it put in the pipeline, with the development of technology, a $4 billion and change a year line, being 10% more efficient. The same transcript records about 130 different projects implemented, each with a payback, none of them huge individual contributors, with overall benefits of 2x to 3x over time of the expense. Anthropic's Mythos is a real model and the access programme is documented on Anthropic's own channels. Anthropic announced Project Glasswing on April 7 2026, an initiative bringing together AWS, Anthropic, Apple, Broadcom, Cisco, CrowdStrike, Google, JPMorganChase, the Linux Foundation, Microsoft, NVIDIA and Palo Alto Networks, formed because of capabilities observed in a new frontier model. Anthropic stated it released the model initially to a limited group of critical industry partners and open source developers, and that it did not plan to make Mythos Preview generally available. Separately, and bearing on the cost context, Bank of America's own SEC filing states that on October 13 2026 the OCC and FinCEN announced the corporation consented to orders related to its Bank Secrecy Act, anti-money laundering and economic sanctions compliance programmes, and that for Q3 2026 the corporation recorded non-tax-deductible noninterest expense of $1.2 billion to accrue for the civil money penalties.
Complete reasoning
The reply is formatted for pasting into the thread where the claim is circulating.
Ask this case
Answers come only from the case file above; nothing is added.
Did Bank of America really hire 1,000 outside contractors to fix AI-exposed software flaws?
Yes, this part checks out. CEO Brian Moynihan said on October 14 2026 that the bank put on about 1,000 contractors to do patching work, working alongside roughly 3,000 existing staff in that area.
When were the contractors actually brought on?
They were put on in the second quarter of 2026, not in October when the statements were made public. Moynihan said those contractors remain in place today.
Was the Q3 spending figure really 'more than $200 million'?
Moynihan actually said the bank spent 'a couple hundred million plus' on remediation, a spoken estimate rather than a precise accounting figure. He also did not break down how much was for AI-related software patching versus anti-money-laundering compliance work.
Does the $200 million figure cover all the bank's AML-related costs that quarter?
No. In the same quarter, Bank of America separately recorded a $1.2 billion non-tax-deductible charge tied to civil money penalties from OCC and FinCEN consent orders over its anti-money-laundering and sanctions compliance programs. This charge is documented in the bank's own SEC filing and is not included in the couple-hundred-million figure.
Do the $700 million in AI savings offset the remediation costs?
Nothing in the transcript connects the two. The $700 million figure refers to projected 2027 benefits from about 130 internal efficiency projects, while the remediation spending is a separate security and compliance cost. The investigation found no statement netting one against the other.